feat(loops): wire analyses → driver — the adaptive-driver channel (Phase 2) by tangletools · Pull Request #135 · tangle-network/agent-runtime

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools · 2026-06-03T15:39:30Z

Roadmap Phase 2 — the analyses → driver wire. This is the load-bearing organ from architecture-interpretations.md: the diagram's red ✗ analyses[] never reaches plan().

Before

PlannerContext carried only {task, history, iterationsSpent, iterationsRemaining} — the driver decided topology from the verdict score alone. runAnalystLoop had zero consumers under src/loops/drivers/.

After — the wire

PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from @tangle-network/agent-eval — substrate type, not redefined).
createDynamicDriver({ analyze }) — an optional hook (AnalyzeInput → findings). plan() calls it each round after the first (a trace must exist), fail-loud on a non-array return (no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule).
renderAnalyses() formats the diagnosis; the canonical sandbox-planner prompt renders it, so the live LLM planner steers from f(trace, findings), not the score. Barrel re-exports AnalyzeInput + renderAnalyses.

Verified

tsc clean; 70/70 loops tests pass. The extended dynamic.test.ts (+4) proves: the hook feeds the planner (and is skipped on round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, and renderAnalyses formatting.

Not in this PR

This is the code wire. Feeding it from a real analyst on a benchmark + the Phase-2 gate (refine@k-with-findings vs random@k at equal k under a non-oracle selector) remain — the gate run is currently blocked on the sandbox gateway↔SDK replay-endpoint skew (see #134 + the recorded blocker). Building the wire now was the explicit call; measuring it follows once the gate is runnable.

…hase 2) Closes the load-bearing gap from docs/architecture-interpretations.md: the driver decided topology from the verdict score alone because PlannerContext had no channel for analyst findings. Now it decides from the diagnosis. - PlannerContext.analyses?: ReadonlyArray<AnalystFinding> (imported from agent-eval — the substrate type, not redefined). - createDynamicDriver gains an optional `analyze` hook (AnalyzeInput -> findings). plan() calls it each round AFTER the first (a trace must exist) and BEFORE the planner, validates the return (fail-loud on non-array — no silent empty), and passes the findings to the planner via ctx.analyses. The hook is the caller's seam to runAnalystLoop, kept on the driver so run-loop stays analyst-free (the layering rule). - renderAnalyses() formats findings (severity/area/claim/recommended_action/confidence); the canonical sandbox-planner prompt renders them so the LLM planner steers from the diagnosis. Barrel re-exports AnalyzeInput + renderAnalyses. - tests/loops/dynamic.test.ts extended (+4): the hook feeds the planner (skipping round 0), the diagnosis renders into the live planner prompt, fail-loud on non-array, renderAnalyses formatting. 70/70 loops tests green; tsc clean. This is the CODE wire (the missing organ). Feeding it from a real analyst on a benchmark + the Phase-2 gate measurement (refine@k-with-findings vs random@k) remain — the latter gated on the sandbox infra (see fix#134 + the gateway replay blocker).

tangletools force-pushed the feat/phase2-analyses-wire branch from 0071719 to fa53764Compare June 3, 2026 15:49

tangletools merged commit c056c5c into mainJun 3, 2026
1 check passed

tangletools deleted the feat/phase2-analyses-wire branch June 3, 2026 15:51

drewstone mentioned this pull request Aug 16, 2026
chore(release): 0.135.4 — accept the agent-interface 0.54 generation and move the sibling floors with it #878
Closed