@kbediako/codex-orchestrator 0.1.35 → 0.1.37

package/templates/codex/AGENTS.md

@@ -1,9 +1,10 @@
-<!-- codex:instruction-stamp 4f9803271a8209cf58746c0a71d87421952a402c884cc0262a8765fa5c456128 -->
+<!-- codex:instruction-stamp 3599cb54f155d730dfc088b0ee6792ebe91e701cc288930064d73eae562b829a -->
 # Agent Instructions (Template)
 
 ## Orchestrator-first workflow
 - Use `codex-orchestrator` pipelines for planning, implementation, validation, and review.
 - Default to `docs-review` before implementation and `implementation-gate` after code changes.
+- Use `docs-relevance-advisory` when you need semantic docs relevance signal without hard-gate behavior.
 - Prefer cloud mode when runs are long-running/parallel and cloud prerequisites are ready.
 - Before cloud mode, verify branch availability, non-interactive setup commands, and required secrets/variables; if missing, run in local `mcp` mode and record why.
 - Before implementation, run a standalone review of the task/spec against the user’s intent and record the approval in the spec + checklist notes.
@@ -49,19 +50,24 @@
   - `P2/P3` findings are tracked follow-ups.
 
 ## Agent role baseline
-- Built-in roles are `default`, `explorer`, and `worker`; `researcher` is user-defined.
+- Built-in roles are `default`, `explorer`, `worker`, and `awaiter`; `researcher` is user-defined.
 - `spawn_agent` defaults to `default` when `agent_type` is omitted; always set `agent_type` explicitly.
 - For symbolic collab runs, prefix spawned prompts with `[agent_type:<role>]` on line one so role intent is auditable from JSONL/manifests.
 - Keep top-level defaults on latest codex by setting `model = "gpt-5.3-codex"` in `~/.codex/config.toml`.
-- Define a user `agents.explorer` role without `config_file` so built-in explorer inherits top-level model defaults.
+- Set `model_reasoning_effort` to at least `high` (CO default: `xhigh`) so spawned agents inherit high reasoning unless role overrides change it.
+- Built-in `explorer` inherits top-level model defaults unless you attach a `config_file`.
 - Spark caveat: `gpt-5.3-codex-spark` is text-only.
-- Use `[agents] max_threads = 8` as the default baseline; raise to `12` only after proving stable tool/runtime behavior.
+- Keep RLM/collab built-ins-first by default; add custom specialist roles only when there is measured value, clear ownership, and validation evidence.
+- Use `[agents] max_threads = 12` with `max_depth = 4` and `max_spawn_depth = 4` as the default multi-agent baseline.
+- Keep fallback usage explicit and rare: `8/2/2` for constrained/high-risk lanes, `6/1/1` only as break-glass.
 - Add an explicit `worker_complex` role (`gpt-5.3-codex`, `xhigh`) for high-risk implementation streams.
+- Use `codex-orchestrator doctor` as an advisory drift check for Codex defaults; remediate additively via `codex-orchestrator codex defaults --yes`.
 
 ## Completion discipline (patience-first)
 - Wait/poll for terminal state on long-running operations (CI checks, reviews, cloud jobs, orchestrator runs) before reporting completion.
 - Reset waiting windows when checks restart or new feedback appears.
 - Do not hand off mid-flight work unless the user explicitly asks to stop.
+- Awaiter triage: treat long waits as expected unless progress is flat across multiple polling windows; increase timeouts before declaring a stall.
 
 ## Instruction stamp
 - If you edit this file, refresh the instruction stamp.