Reconcile the diverged agent-improver definitions into one generic source + deployment shim

[devantler]

🤖 Generated by the Daily AI Engineer

First child of #76. The shipped plugins/automated-ai-engineer/agents/agent-improver.md (154 lines) and the deployment's live .claude/agents/agent-improver.md in devantler-tech/monorepo (170 lines) have diverged — the local copy is the one that runs daily, so its improvements never reach the product.

Scope

• Diff the two; fold every generic improvement from the local copy into the shipped definition, and every deployment-specific detail (telemetry paths, both-instance specifics, the symmetric-authority grant, schedule) into the consuming deployment's configuration surface.
• Reduce the monorepo copy to the thinnest working shim over the shipped definition — or delete it if the plugin-loaded agent verifiably serves the scheduled run.
• Add the drift guard: CI or structure that prevents silent re-forking (the agent-files-stay-in-sync rule, enforced).

Acceptance criteria

• No generic prose exists only in the monorepo copy (verified by diff, recorded on the PR).
• The daily 06:30 run demonstrably still works after the convergence (next run's report cited).
• Authority in the shipped definition reads default-gated; the deployment's wider grant lives only in deployment config.
• Drift between the two surfaces fails CI or is structurally impossible.

Rough size: S–M. Behaviour-preserving for the deployment; never mixed with capability changes.

[coderabbitai]

🔗 Related PRs

#34 - docs: describe the live multi-upstream skill-sourcing model [merged]
#36 - feat: bundle in-house agent-skills into category plugins [merged]
#53 - fix: omit skills/agents path fields so plugins install in Claude Code [merged]
#55 - docs(adr): define the Automated AI Engineer plugin boundary [merged]
#63 - feat: add the automated-ai-engineer plugin (agents+skills bundle per ADR-0002) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[devantler]

🤖 Generated by the Agentic Engineer

The definition convergence and delivery contract are merged:

• upstream devantler-tech/agent-plugins#86 → f5a4ce490cde3dcc79cd2e4a744a07fdbefd1905
• consumer devantler-tech/monorepo#2404 → 6ecf5308ac7f4f926d12207a5cf04cecad28084c

The monorepo's local Agent Improver agent/skill forks are deleted, generic behavior is plugin-owned, deployment-specific symmetric authority and writer-lane mapping live in AGENTS.md, and CI fails closed on drift or loss of delivery semantics. Both PRs merged only after exact-head CI, effect evaluation, clean Codex review, and resolved threads.

Issue #77 stays open for its final post-deployment acceptance check: verify and cite the first scheduled 06:30 Europe/Copenhagen Agent Improver run after convergence, due 2026-07-24. Until that report proves the installed schedule loads and executes the merged source, the runtime verification hypothesis remains open.

[devantler]

🤖 Generated by the Agentic Engineer

Post-deployment acceptance is complete in the 2026-07-25 Codex Agent Improver automation run.

The run loaded the thin Codex schedule pointer, native automation memory, the canonical consumer AGENTS.md, the installed agentic-engineering cache, and latest reviewed upstream main before invoking agent-improver. The entrypoint SHA-256 is identical in all three deployed/reviewed locations:

d4b6a57429e80d0334866a9d354cb1e11fd6303df694132eb34a601d45689498

• installed cache agentic-engineering/2.0.0/agents/agent-improver.agent.md
• consumer gitlink source at 0c3c3ff
• latest reviewed upstream source at 567459d

The consumer's named definition-location, authority, memory, maintainer-channel, and writer-namespace facts all resolved, so the role did not fail closed. It completed the live evidence loop, claimed agent-plugins#96, and opened agent-plugins#97 with RED/GREEN contract proof. This satisfies the remaining scheduled-run verification criterion.