Skill

aidd-stage-research

Defines stage rules for researcher subagent in AIDD/RLM pipeline: evidence reading from packs/worklists, targeted slicing via rlm_slice.py, handoffs to rlm_finalize.py when blocked.

Python

ai-ml

automation

npx claudepluginhub grinrus/ai_driven_dev --plugin feature-dev-aidd

Configuration

Model: inherit

Tool Access

This skill uses the workspace's default tool permissions.

Preview

- Preload-only reference for `feature-dev-aidd:researcher` subagent.

SKILL.md

Similar Skills

gsd:research-phase

Researches technical implementation approaches for a project phase by validating input, checking existing research, gathering context, and spawning a gsd-phase-researcher agent. Standalone; prefer /gsd:plan-phase.

3 tools

gsd

research-plan-implement

Orchestrates end-to-end Research-Plan-Implement pipeline using parallel subagents per phase with file artifacts for communication. Use for complex features needing codebase research, planning, and implementation in one session.

rpikit

Research Execute Skill

Executes research code in src/ to generate artifacts in results/, reading commands from plan/research_plan.md YAML frontmatter or execution_manifest.json. Phase 3.5 of research pipeline with prerequisite checks.

magi-researchers

Stats

Stars5

Forks0

Last CommitMar 6, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Scope

Preload-only reference for feature-dev-aidd:researcher subagent.
Defines stage behavior for research content work in the RLM-only pipeline.
Keeps stage rules separate from the user-invocable researcher command skill.

Stage contract

skills/researcher/SKILL.md owns orchestration only (state sync, pipeline run, explicit subagent call, handoff).
agents/researcher.md owns content updates (aidd/docs/research/<ticket>.md) based on existing evidence.
Subagent must not run shared RLM owner internals directly (rlm_nodes_build.py, rlm_links_build.py, rlm_finalize.py).
Stage runtime may run bounded canonical finalize recovery once in --auto mode; manual recovery paths remain outside subagent scope.

Command contracts

`python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_slice.py`

When to run: for bounded evidence lookup during researcher subagent analysis.
Inputs: ticket/query tuple for targeted code-artifact context extraction.
Outputs: slice evidence blocks suitable for research document updates.
Failure mode: non-zero exit on missing pack/worklist artifacts or invalid query.
Next action: surface blocker and handoff to RLM owner flow, then rerun slice when artifacts are ready.

`python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_finalize.py`

When to run: handoff-only path when stage runtime reports pending RLM readiness after bounded auto recovery.
Inputs: --ticket <ticket> from deterministic pending output.
Outputs: finalized RLM readiness state (ready or pending with reason code).
Failure mode: non-zero exit when upstream targets/manifest/worklist integrity is broken.
Next action: resolve upstream integrity issues and rerun finalize via shared owner flow.

Evidence policy

Read pack-first: aidd/reports/research/<ticket>-rlm.pack.json.
If pack is missing, read *-rlm.worklist.pack.json and return BLOCKED with explicit handoff.
Use slice for targeted evidence only:
- python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_slice.py --ticket <ticket> --query "<token>".
Avoid full JSONL scans unless pack/worklist is insufficient.

Handoff policy

When rlm_status is not ready, return BLOCKED with deterministic next action:
- python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-rlm/runtime/rlm_finalize.py --ticket <ticket>.
Pending output must include deterministic reason_code + next_action and append research handoff tasks.
When evidence is ready, update research document and return READY with next stage command hint.

Additional resources

Stage command owner: ../researcher/SKILL.md (when: researcher stage orchestration order is unclear; why: keep stage entrypoint behavior aligned with no-fork command contract).
Shared RLM owner: ../aidd-rlm/SKILL.md (when: pending/finalize ownership boundaries are unclear; why: enforce explicit handoff from stage to shared RLM runtime owner).