docs: hand off Lousy Agents research into Agent OS (docs only) - #92
Conversation
Split the two decision-record docs out of the draft experiment branch (PR #83) so the settled research lands on main while the paid experiment harness stays behind issue #85's do-not-merge guardrail. Review fixes applied in place: - lousy-agents-handoff.md: reword the redgate suggested-skills line to the merged #82 boundary — compose redgate around the applicable specialist; it reinforces execution, it does not own the domain procedure. - automation-curation-lens.md: sync the Redgate framing to #82's merged vocabulary (working harness/protocol layer, verification/gating envelope, specialist-first). - automation-curation-lens.md: add the missing multiplicity legend for the relationship diagram (unresolved Copilot review finding on Adapter+). - automation-curation-lens.md: pin the DESIGN_EVIDENCE.md link to the experiment branch commit, since the harness does not land with this docs-only change. - verified no stale .Agent tokens remain in either doc. Part of #83; relates to #84 #85. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NsPQTo67oovz8ogirsH6uL
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6075ecfaba
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| -> contains Workstream | ||
| -> contains Automation | ||
| -> triggeredBy Trigger | ||
| -> follows Recipe* |
There was a problem hiding this comment.
The canonical graph permits zero Recipe bindings via follows Recipe*, but the adjacent Automation definition requires “one or more recipes” (line 115). Implementing validation or adapters from this decision record can therefore produce incompatible schemas for recipe-less jobs; choose the intended cardinality and use it consistently.
Useful? React with 👍 / 👎.
| [workflow run](https://github.com/JRichlen/agent-plugins/actions/runs/33281138920) | ||
| and [raw artifact](https://github.com/JRichlen/agent-plugins/actions/runs/33281138920/artifacts/9723030558) | ||
| are the evidence authority. |
There was a problem hiding this comment.
The document declares this Actions artifact to be the evidence authority even though GitHub Actions artifacts expire according to repository retention settings. Once that happens, readers cannot inspect the raw evidence behind these placement decisions—especially because the experiment harness is intentionally absent from this tree—so preserve the artifact in a commit or another durable location rather than relying on the run artifact URL.
Useful? React with 👍 / 👎.
…ing.md per the standing order The run.sh sections auto-merged cleanly. Per the standing order that #91 landed, this PR re-scopes the routing tier (typed composition line + the trajectory leg), so docs/testing.md's routing entry now documents both legs, the two contract validators, the offline tests, the second pass-rate gate, and the updated trigger paths and call counts. Gates on the merged tree: cheap 1512/0, counterfeits 22/0, branch protection and behavior-surfaces lockstep in sync. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NsPQTo67oovz8ogirsH6uL
…ra 7) Daily refresh: main advanced with #92, which lands the automation-curation lens as an explicit decision record. New era-7 entry with receipts to the PR, the decision record, the handoff that posed the question, and the docs commit: Agent OS as a design/control plane one semantic layer above harness configuration, Redgate deliberately a different (optional execution-policy) layer, a seven-concept ontology instead of imported harness constructs, prior art used as prior art rather than a fork target, and per-capability evidence-rated harness support. Placed in era 7 rather than opening an era: it extends the boundary story from the side above the envelope, and one docs-only PR does not clearly open a new movement — the owner's curation pass can promote it. Also merges origin/main (the PR was behind). Timeline guard in sync; cheap tier 1183/0. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014XYwiWZicXu2CfDUKAXHpi
What
The docs half of #83, split out per review so the settled research can land on main while the paid experiment harness stays on the original branch under #85's do-not-merge guardrail.
Two new decision-record documents under
docs/research/agent-os/:lousy-agents-handoff.md— the pointer-first handoff of thezpratt/lousy-agentsresearch: prior-art seams to inspect, the projection/adapter architectural implication to test, the receiving-thread lenses, and suggested skills for the work.automation-curation-lens.md— the receiving-thread decision record for automation taxonomy/curation: the lane-scoped naming invariant, the seven-concept minimal ontology, Recipe-vs-Automation distinctions, the Agent OS / Redgate layer model, theGov/Metaboundary, three-truths reconciliation, the meta-curation loop, progressive-disclosure recipe structure, and the adapter capability matrix.Nothing from
experiments/or.github/comes over. #83 continues to carry the experiment harness and workflows only.Doc fixes applied in place
a.
lousy-agents-handoff.md— redgate suggested-skills line reworded to the merged #82 boundary. The old line read as routing to redgate as a domain specialist.Before:
After:
b.
automation-curation-lens.md— Redgate framing synced to #82's merged vocabulary. The Decision section's "Redgate is a harness/protocol used while doing nontrivial work" paragraph now uses the merged wording: a working harness/protocol layer — a verification/gating envelope composed around the most-specific applicable specialist, not the universal router, while the specialist owns the domain procedure. Light touch; the rest of the doc (layer model, plugin boundaries) was already consistent.c. Stale
.Agenttokens. Checked both docs, including the naming-convention line<Lane> <Workstream>.<Automation>: <Short Name>— the rename had already fully landed; no stragglers found, no change needed.d.
automation-curation-lens.md— missing multiplicity legend forAdapter+(unresolved Copilot review finding on #83). A one-line legend now follows the relationship diagram, matching the doc's existing suffix style.Before (diagram excerpt, no legend):
After (legend added directly below the diagram):
Additionally, the relative link to
experiments/agent-os/DESIGN_EVIDENCE.mdwould have been dangling on main after this docs-only merge (the harness stays on #83's branch), so it is now pinned to that branch's commit on GitHub with a note.Gates
evals/cheap/run.sh: exit 0 — 1178 passed, 0 failedpython3 ci/check_branch_protection.py --repo .: exit 0 — 6 checks + 2 safety paths in syncPart of #83 / relates to #84 #85.
🤖 Generated with Claude Code
Generated by Claude Code