Literature Review

Write

• projects/<slug>/workspace/bootstrap/literature-review.md
• projects/<slug>/STATE.md

Must Not Write

• projects/<slug>/experiment-memory.md
• projects/<slug>/review-state.json
• projects/<slug>/results.tsv
• projects/<slug>/plans/<idea>/anchor.md
• projects/<slug>/decision-tree.md
• memory/*
• projects/<slug>/workspace/dashboard-data.json

Required Input Checks

Before proceeding, verify:

• project-brief.md exists and is structurally readable
• intake_mode is either direction-search or seed-papers
• target_problem is non-empty
• at least one of these is present:
  • a usable direction_prompt
  • one or more seed_papers

If the intake is incomplete, stop and report the exact missing field instead of fabricating a search brief.

Protocol

1. Load Intake Context

Read:

• project-brief.md for research goal, intake mode, constraints, venue targets, execution resources, and integrity red lines
• STATE.md for the current project phase and bootstrap posture
• relevant recurring patterns from memory/lessons-learned.md and memory/failure-library.md

2. Choose The Review Mode

If intake_mode: direction-search:

• expand from the target problem into problem families, baseline clusters, evaluation norms, and likely dead ends

If intake_mode: seed-papers:

• center the review on the supplied seed papers
• map nearest-neighbor work, claimed differentiators, reused evaluation setups, and likely novelty traps

3. Build A Conservative Opportunity Map

Write projects/<slug>/workspace/bootstrap/literature-review.md using the documented bootstrap contract.

The artifact should capture:

• the scoped search frame
• the most relevant papers or artifacts reviewed
• candidate opportunity zones worth ideation
• repeated failure signals or anti-patterns drawn from global memory
• unresolved questions that still need idea-level narrowing

Keep the artifact evidence-oriented. Do not claim a definitive novelty gap unless the evidence supports it.

4. Update Project Steering State

Update STATE.md conservatively:

• set phase: phase1 if the project was still in phase0
• set project_status: running
• keep active_idea_id: null
• keep active_branch_id: null
• set next_action: generate bounded candidate ideas from the literature review artifact
• set decision_mode: auto-report
• set human_attention: none
• clear decision_type and decision_options_ref
• refresh risk_level and blockers only if the review surfaced a real constraint
• set last_completed_skill: literature-review
• refresh last_updated

Failure Handling

Stop without writing if:

• the intake mode is unreadable or unsupported
• the project brief does not provide enough direction to scope a review
• the requested review would require inventing sources or hidden state

When stopping, report the exact missing field or contract gap.

Self-Check

• Wrote only the bootstrap literature artifact plus steering updates in STATE.md
• Did not create an active experiment line
• Reused lessons and failure patterns by reference rather than copying global memory into canonical project state
• Kept novelty claims conservative and traceable
• Left experiment-memory.md, results.tsv, and review-state.json untouched