researcher

name	researcher
description	Generates or refreshes research artifacts and RLM evidence for a ticket. Use when research stage should produce or update canonical RLM outputs. Do not use when the request is initial ticket bootstrap in `idea-new` or implementation planning in `plan-new`.
argument-hint	$1 [note...] [--paths path1,path2] [--keywords kw1,kw2] [--note text]
lang	en
prompt_version	1.2.35
source_version	1.2.35
allowed-tools	["Read","Edit","Write","Glob","Bash(rg *)","Bash(sed *)","Bash(python3 ${CLAUDE_PLUGIN_ROOT}/skills/researcher/runtime/research.py *)","Bash(python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/set_active_feature.py *)","Bash(python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/set_active_stage.py *)","Bash(python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/tasks_derive.py *)"]
model	inherit
disable-model-invocation	true
user-invocable	true

Follow feature-dev-aidd:aidd-core.

Steps

1. Set active feature and stage research.
2. Run python3 ${CLAUDE_PLUGIN_ROOT}/skills/researcher/runtime/research.py --ticket <ticket> --auto.
3. Re-run the same entrypoint with optional overrides (--paths, --keywords, --note) when targeted refresh is needed.
4. Validate RLM outputs (*-rlm-targets.json, *-rlm-manifest.json, *-rlm.worklist.pack.json, optional *-rlm.pack.json).
5. Run subagent feature-dev-aidd:researcher. First action: read RLM pack/worklist.
6. In --auto mode run bounded canonical finalize recovery once (rlm_finalize --bootstrap-if-missing) from stage runtime.
7. If RLM is still pending after auto recovery, return deterministic pending reason + explicit next action (python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_finalize.py --ticket <ticket>) and append research handoff via python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/tasks_derive.py --source research --append.
8. Ready path: return /feature-dev-aidd:plan-new <ticket> only when RLM readiness is ready and the research document is synchronized with the current evidence.
9. Pending or blocked path: if readiness is not ready, return the deterministic blocker or pending output instead of guessing a downstream handoff.
10. Runtime-path safety: execute only canonical runtime commands from this contract (python3 ${CLAUDE_PLUGIN_ROOT}/skills/.../runtime/...).
11. Root-relative /skills/... runtime paths are forbidden; use only ${CLAUDE_PLUGIN_ROOT}/skills/.../runtime/....

Command contracts

python3 ${CLAUDE_PLUGIN_ROOT}/skills/researcher/runtime/research.py

• When to run: always as canonical researcher pipeline entrypoint.
• Inputs: --ticket <ticket> with optional path/keyword/note overrides.
• Outputs: research artifacts, stage status, and RLM readiness/handoff markers (rlm_status, rlm_pack_status).
• Failure mode: non-zero exit when required inputs/artifacts are missing or RLM artifact generation fails.
• Next action: fix missing inputs/paths, rerun this entrypoint; in auto mode finalize/bridge runs once, then pending outcomes include deterministic reason_code + next_action.

Notes

• Planning stage: AIDD:ACTIONS_LOG: n/a.

Migration policy

• Legacy pre-RLM research context/targets artifacts are ignored by runtime/gates.
• For old workspace state, rerun: python3 ${CLAUDE_PLUGIN_ROOT}/skills/researcher/runtime/research.py --ticket <ticket> --auto.
• If rlm_status remains pending after bounded auto recovery, hand off to shared owner: python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_finalize.py --ticket <ticket>.

Additional resources

• Research template source: templates/research.template.md (when: drafting or reviewing aidd/docs/research/<ticket>.md; why: keep artifact shape aligned with canonical workspace template).
• Shared RLM owner skill: ../aidd-rlm/SKILL.md (when: rlm_status is not ready or RLM pack is missing; why: run canonical shared RLM API outside stage-local orchestration).