feat(loops): recursive execution atom — budget-conserving Scope + Supervisor keystone by drewstone · Pull Request #151 · tangle-network/agent-runtime

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone · 2026-06-04T12:34:22Z

What

The v1 recursive execution atom — the keystone for drivers-of-drivers: one self-similar Agent whose act spawns child agents through a Scope, run by a Supervisor that owns a conserved budget pool, an event-sourced journal, and an observability/conversation handle. The flat experiment harness is recovered as the simplest act (it does not compete with this — it is one program over it).

Design + decision record: docs/research/recursive-execution-atom.md (frozen contract, build order, the 4 resolved forks, the adversarial critique that shaped the surface).

The pieces

src/loops/supervise/types.ts — the frozen contract. Agent, an open LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, a user's own agent is first-class via the registry/BYO — no per-vendor adapters), Scope, Supervisor, Settled, Budget.
supervise/budget.ts — a conserved reservation pool: atomic reserve-on-spawn, fail-closed admission, refund-on-settle. This is the load-bearing invariant: Σk(treatment) ≡ Σk(blind) holds by construction, so a steered arm can never silently out-compute blind (the confound that burned the earlier "+20pp steering" result).
src/durable/spawn-journal.ts — event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay. Resumable, queryable, reproducible from one log.
supervise/scope.ts — a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled → Iteration adapter keeps defaultSelectWinner single-sourced.
supervise/runtime.ts — the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork.
supervise/supervisor.ts — nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate.
bench/src/drivers/ — flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. WidenGate defaults to flat so the selector≠judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt.
program.ts — mapPool one-for-one failure semantics (a down child is excluded from the merge; an all-down batch re-throws the first original error so a maxDepth guard still propagates loud).

Verification

Independently re-run on this branch: typecheck ✓, lint ✓ (204 files), build ✓, full suite 642/642 (incl. 28 keystone property tests: conserved-budget fail-closed + refund, equal-k by construction, monotonic-seq cursor, abort/teardown, replay determinism), bench typecheck ✓.

Honest gaps (why draft)

Live executor paths unproven. The router/sandbox/cli executors are exercised only through the offline mock LeafExecutor — they typecheck and the wiring is proven, but no real router-HTTP / sandbox-runLoop / cli-subprocess run has happened.
The LLM-meta-driver is the gated treatment, not a default. It ships so the diverse-strategy-vs-blind gate can actually be run; it is not wired into any default path.
Deferred (gated on a positive gate result): a tuned MCTS-PW algorithm, learned widening, a Temporal/DBOS durable backend, deleting runProgram's loop-layer parallel op.

Relationship to #150

#150 adds the leaf-level continued-session/fork lineage on runLoop. This is the driver layer on top — the sandbox executor forwards that lineage passthrough rather than duplicating it. #150's findings (esp. verify the client-minted sessionId) are load-bearing here too.

Capture the design thread as tracked research docs under docs/research: - recursive-execution-atom: the next generation (one recursive Agent atom run as a durable, observable supervision tree; analyst-as-agent-with-runtime; async dynamic spawning), the proposed surface, the file-grounded gap, and the open forks. Plane B contains the flat harness. - flat-harness-design: the assumption-free experiment harness synthesis (profiles x steer x executionMode x allocation). Plane A. - long-horizon-benchmark-survey: adversarially-verified survey; Commit0 and tau2-bench as the multi-turn picks. Index them under the new Research track in docs/README.md.

…spec Freeze the contract: the budget-conserving reactive Scope + Supervisor keystone, the event-sourced SpawnJournal + ResultBlobStore (outRef replay), the LeafExecutor per-harness model (harness:null = Router inference, sandbox, cli), and the 8-step v1 build order. Records the 4 resolved forks and the operator override (build the LLM meta-driver now as the treatment on top of the conserved reservation pool, so the equal-k gate stays valid by construction).

Operator refinement: the runtime is ONE open interface with an execute that returns a promise or an async stream, not a closed inline|sandbox|cli union. Built-ins are implementations; a user agent (mastra/agno/HTTP/custom) is first-class by implementing it; no per-vendor adapters. The sandbox executor composes runLoop and forwards PR #150's lineage passthrough rather than reinventing checkpoint/fork. Records the #150 review (approve-to-land; verify client-minted sessionId, bound fork acquisition, document parent-image fork).

…ervisor The v1 keystone for the recursive-agent-atom (drivers-of-drivers, async, observable). One self-similar atom whose act spawns child agents through a Scope; the flat experiment harness is recovered as the simplest act. - src/loops/supervise/types.ts: the frozen contract — Agent, an OPEN LeafExecutor interface (execute returns a promise or an async stream; router/inline + sandbox + cli are implementations, BYO is first-class via the registry; no per-vendor adapters), Scope, Supervisor, Settled, Budget. - supervise/budget.ts: a conserved reservation pool — atomic reserve-on-spawn, fail-closed admission, refund-on-settle, so equal-k holds by construction (the invariant that keeps a steered arm from silently out-computing blind). - durable/spawn-journal.ts: event-sourced SpawnJournal + content-addressed ResultBlobStore + seq-ordered replay (resumable, queryable, reproducible). - supervise/scope.ts: a ray.wait cursor over an in-memory nursery; spawn reserves budget and resolves the executor through the open registry; a Settled-to-Iteration adapter keeps defaultSelectWinner single-sourced. - supervise/runtime.ts: the open executor registry; the sandbox executor composes runLoop (forwarding an optional lineage passthrough) rather than reinventing checkpoint/fork. - supervise/supervisor.ts: nursery join barrier, abort cascade (incl. the acquire lifecycle), OTP intensity breaker, typed SupervisedResult, RootHandle (view/signal/abort) as the observability substrate. - bench/src/drivers: flat-harness control (with the equal-k assertion), progressive-widening control, and the LLM-meta-driver treatment. The WidenGate defaults to flat so the selector-not-judge firewall stays dormant; widening reads trace findings, never the raw verdict, unless judgeExempt. - program.ts: mapPool one-for-one failure semantics (a down child is excluded from the merge, an all-down batch re-throws the first original error so a maxDepth guard still propagates loud). Verified: typecheck, lint (204 files), build, full suite (642 tests incl. 28 keystone property tests), bench typecheck. Caveat: the live executor paths (real router HTTP, sandbox runLoop, cli subprocess) are exercised only through the offline mock LeafExecutor; no live-backend run yet.

tangletools · 2026-06-04T13:03:47Z

🔍 Reviewing `78d80209`

Pass	Status	ETA
opencode DeepSeek v4 Pro	Running (2 min)	~5-15 min
opencode GLM 5.1	Running (2 min)	~5-15 min

Agent review running. Reads the actual code. This comment updates in place.

_{tangletools · #151 · model: kimi-for-coding · started 2026-06-04T13:05:48Z}

…ion-atom # Conflicts: # src/loops/index.ts

Cuts the 58-commit backlog on main into a published release. Headline surface: - runToolLoop / streamToolLoop — bounded turn-level tool-dispatch loop (#137) - RSI agent tree: recursive Agent.act, Supervisor keystone, runProgram, the adaptive-driver channel (#139/#151/#165) - optimization API collapsed onto agent-eval selfImprove; the runtime keeps the CODE-surface ImprovementDriver you pass as driver (#172) - deployable benchmark adapters: AppWorld, commit0, aec-bench, EnterpriseOps-Gym; runBenchmarks over one ADAPTERS registry (#153/#156/#157) - agent-eval floor raised to >=0.83.0 (#175)

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

drewstone added 4 commits June 4, 2026 05:26

drewstone marked this pull request as ready for review June 4, 2026 13:02

Merge remote-tracking branch 'origin/main' into feat/recursive-execut…
e0fbd3b
…ion-atom # Conflicts: # src/loops/index.ts

drewstone merged commit 06efe71 into mainJun 4, 2026
1 check passed

drewstone deleted the feat/recursive-execution-atom branch June 4, 2026 13:06

drewstone mentioned this pull request Jun 6, 2026
chore(release): agent-runtime 0.45.0 #176
Merged

drewstone mentioned this pull request Jun 6, 2026
chore(hygiene): fix stale CLAUDE.md code-map; (wip) de-dup steer-firewall + bench stats #183
Merged

Uh oh!

Conversation

drewstone commented Jun 4, 2026

What

The pieces

Verification

Honest gaps (why draft)

Relationship to #150

Uh oh!

tangletools commented Jun 4, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 Reviewing 78d80209

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 4, 2026

What

The pieces

Verification

Honest gaps (why draft)

Relationship to #150

Uh oh!

tangletools commented Jun 4, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 Reviewing 78d80209

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 4, 2026

What

The pieces

Verification

Honest gaps (why draft)

Relationship to #150

Uh oh!

tangletools commented Jun 4, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 Reviewing 78d80209

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

drewstone commented Jun 4, 2026

What

The pieces

Verification

Honest gaps (why draft)

Relationship to #150

Uh oh!

tangletools commented Jun 4, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔍 Reviewing 78d80209

Uh oh!

Uh oh!

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`

tangletools commented Jun 4, 2026•
edited
Loading

🔍 Reviewing `78d80209`