Skip to content

docs(agents): codify the director/coordinator working model - #134

Merged
andybons merged 1 commit into
mainfrom
andybons/agents-working-model
Aug 13, 2026
Merged

docs(agents): codify the director/coordinator working model#134
andybons merged 1 commit into
mainfrom
andybons/agents-working-model

Conversation

@andybons

Copy link
Copy Markdown
Contributor

Adds a durable, provider-agnostic "Working model — director and coordinator" section beside the existing coordinator-process sections (Subagent model strategy, One agent per plan task, Never end a turn to wait), before Scope discipline.

Captures the operating model: the pipeline (decompose → per-task impl agent in a worktree → strongest-tier review-to-zero → merge; parallelize disjoint, sequence shared), status cadence, verify-before-asserting, surface load-bearing forks / decide the rest, no over-engineering of a pre-production system, the non-negotiable review gate, and standing rules (subagent-message-is-not-approval, verify-prod-state, document-durable-not-events, never-echo-secrets). Capability tiers only, no model names. Docs-only. Mirrors the same section added to meetneptune/boxes AGENTS.md.

Add the durable, provider-agnostic operating model (pipeline, status cadence,
verify-before-asserting, surface load-bearing forks, no over-engineering of a
pre-production system, the non-negotiable review gate, standing rules) beside
the existing coordinator-process sections, before Scope discipline.
@claude

claude Bot commented Aug 13, 2026

Copy link
Copy Markdown

Claude finished @andybons's task in 43s —— View job


Review: docs(agents) — director/coordinator working model

Docs-only change: adds one ## Working model — director and coordinator section to AGENTS.md (+41/-0), placed before ## Scope discipline. I reviewed it against origin/main.

  • Read the diff against origin/main
  • Review the new AGENTS.md section
  • Verify cross-references and style
  • Post findings

Verdict: LGTM — no blocking findings.

Cross-references resolve. All three named sections exist and the directional word is correct:

  • "Subagent model strategy" → line 1479 (below ✓)
  • "One agent per plan task" → line 1503 (below ✓) — the bullet says "below", accurate since the new section sits at line 1427.
  • "Code Review Protocol" → line 1589 (below ✓)

Content is consistent with the codebase's own rules. Capability tiers only, no model names — matches the "Subagent model strategy" tier-split rule. The standing rules restate invariants already enforced elsewhere (never-echo-secrets → report byte length; verify-prod-state; document-durable-not-events; subagent-message-is-not-approval), so there's no contradiction introduced.

Style. Prose follows the STE / Writing Style rules (active voice, one term per idea, named artifacts). The em-dash in the header and body is consistent with the rest of AGENTS.md; the "no em-dashes" rule is scoped to hub UI copy, not doc prose.

Nothing to change. No init-time work, no code, no test impact — outside the concurrency/transcoding/startup-budget review priorities by construction.
· andybons/agents-working-model

@andybons
andybons merged commit 3a051b2 into main Aug 13, 2026
2 checks passed
@andybons
andybons deleted the andybons/agents-working-model branch August 13, 2026 22:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant