Checks specific claims and mechanisms in a research plan iteration against the academic literature and updates the threat map in place. Use when inside a planning loop and the research plan has been created or revised, and the threat map needs to be checked against the new plan content. Also use when asked to do a "targeted literature check", "plan-specific novelty check", or "update the threat map for this iteration". Do NOT use for initial broad scans (use literature-review-light instead) or for final exhaustive reviews (use literature-review-deep instead).
Check whether the current iteration of the research plan introduces claims, mechanisms, or contributions that are at risk given the existing literature. Update the threat map in place. Do not repeat work already done — focus only on what is new or changed since the last iteration.
This skill runs inside the planning loop, potentially up to 5 times. Be efficient. Every search should have a reason.
Calibrate all output to a PhD / top-journal standard. Cite by author-year. Reference specific propositions, theorems, and mechanisms.
Read all of these before doing anything else:
context/planning/research_plan.md — the current iteration's plan (check the changelog at the bottom to identify what changed this iteration)context/literature/threat_map.md — the existing threat map (this is the file you will update)context/research_context.md — the permanent project specificationcontext/literature/constraints.md — if present, constraints on what the literature has and has not addressedcontext/literature/search_log.md — if present, log of prior searches (read to avoid re-searching)context/evaluator_feedback.md — if present, evaluator's literature-related concerns from the previous scoring roundIdentify what is new or changed in this iteration of research_plan.md. Look for:
evaluator_feedback.md exists)If the plan has a changelog at the bottom, start there. If not, compare the current plan against what the threat map already covers.
Produce a short internal list of check targets: the specific claims or mechanisms that need literature verification this iteration. If nothing substantive changed, state this explicitly and skip to Step 4 (changelog-only update).
For each check target, search the literature. Use web search to find papers that could overlap with the new or revised claims.
Search strategy:
Do not re-search for papers already in the threat map unless the plan revision changes the overlap assessment.
For each new paper found, classify it using the threat level definitions and paper entry schema in .claude/skills/literature-review/shared-schemas.md.
Additionally, reassess any existing threat map entries that are affected by the plan revision:
[DEPRIORITISED] but not deletedUpdate context/literature/threat_map.md in place. Do not rewrite the entire file. Make only these changes:
---
## Changelog
### Iteration [N] — [YYYY-MM-DD]
**Plan changes checked:**
- [brief description of each check target]
**New papers added:**
- [Author (Year)] → [Channel X] → [Threat level]
- ...
**Reclassifications:**
- [Author (Year)]: [OLD level] → [NEW level]. Reason: [why]
- ...
**No-change confirmation:**
- [List any check targets where no new threats were found]
**Assessment change:** [Yes/No. If yes, explain how the overall threat picture shifted.]
Files updated in place:
context/literature/threat_map.md — updated with changelog entry for this iterationcontext/literature/constraints.md — updated if new constraints are discoveredcontext/literature/search_log.md — append new searches and papers reviewed this iterationThe targeted review does not write literature notes or reviews — those are deep review outputs.
The update must:
[UNVERIFIED]Do not:
literature_notes.md, no mechanism_map.md)literature_constraints.md if it exists — it records what has already been ruled out