Subsection Briefs

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass, no license.

.claude/skills

h = int(hashlib.md5(seed.encode("utf-8", errors="ignore")).hexdigest()[:8], 16)

Audit 0.4.1 · source sha 81994516a430 · 19 Oct 2026Report an issue or request a re-scan

• PublisherWILLOSCAR GitHub userOwner of the repository this skill is read from
• Source folderWILLOSCAR/research-units-pipeline-skills/.codex/skills/subsection-briefsThe exact folder we audited
• LicenceNot declaredAs the repository states it
• Commit pinned for this verdict81994516a430A new audit runs when the source changes
• Repository statusNot archivedLast push 14 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Build per-subsection writing briefs (NO PROSE) so later drafting is driven by evidence and checkable comparison axes (not outline placeholders).

---
name: subsection-briefs
description: "Build per-subsection writing briefs (NO PROSE) so later drafting is driven by evidence and checkable comparison axes (not outline placeholders)."
---

# Subsection Briefs

## Triggers & routing

- **Trigger**: subsection briefs, writing cards, intent cards, H3 briefs, scope_rule, axes, clusters, 写作意图卡, 小节卡片, 段落计划.


`outline/subsection_briefs.refined.ok` freezes reviewed briefs only while the marker is newer than the briefs, declared inputs, domain packs, and generator. Stale markers are removed before backed-up regeneration.

Build deterministic H3 brief cards from outline + mapping + paper notes.

Compatibility mode is active: this skill keeps the current `outline/subsection_briefs.jsonl` field contract and paragraph-plan shape while moving phrase/domain logic into `references/` and `assets/`.

## Quick Use

- Run `scripts/run.py` as the deterministic materializer.
- Keep the output NO PROSE: subsection-scoped plans, axes, clusters, and bridge handles only.
- Preserve current downstream compatibility for `transition-weaver`, `writer-context-pack`, and `subsection-writer`.

## Load Order

Always read:
- `references/overview.md`

Read by task:
- If `thesis` feels repetitive or copyable, read `references/thesis_patterns.md`.
- If `tension_statement` is too generic, read `references/tension_patterns.md`.
- If axes are weak or domain-biased, read `references/axis_catalog_generic.md` and `references/axis_catalog_llm_agents.md`.
- If transition handles feel bland, read `references/bridge_terms.md`.
- For calibration, read `references/examples_good.md`.

Machine-readable assets:

- `assets/phrase_packs/thesis_patterns.json`
- `assets/phrase_packs/bridge_contrast.json`
- `assets/domain_packs/generic.json`
- `assets/domain_packs/llm_agents.json`
- `assets/domain_packs/embodied_ai.json`
- `assets/domain_packs/rag_evaluation.json`
- `assets/domain_packs/text_to_image.json`

The script loads these packs first; patch them before changing Python when the issue is phrasing, domain routing, axis inventory, cluster purity, or lexical bridge coverage.

## Inputs

- `outline/outline.yml`
- `outline/mapping.tsv`
- `papers/paper_notes.jsonl`
- Optional: `GOAL.md`
- Optional: `outline/claim_evidence_matrix.md`

## Output

- `outline/subsection_briefs.jsonl`

Required record shape remains compatibility-preserving:

- identity: `sub_id`, `title`, `section_id`, `section_title`
- planning core: `rq`, `thesis`, `scope_rule`, `axes`, `bridge_terms`, `contrast_hook`, `tension_statement`
- evidence hooks: `evaluation_anchor_minimal`, `required_evidence_fields`, `clusters`
- execution plan: `paragraph_plan`, `evidence_level_summary`, `generated_at`

## What `run.py` Should Do

- Read outline, mapping, and notes.
- Normalize subsection seeds from outline bullets.
- Load thesis/tension/domain-axis packs from `assets/`.
- Produce stable JSONL records with the existing contract.

## What `run.py` Should Not Do

- Do not invent papers, citations, or claims.
- Do not emit reader-facing narrative prose.
- Do not hardcode domain-specific sentence templates when an asset pack can hold them.

## Block / Reroute

- If outline, mapping, or notes are missing, stop.
- If evidence is thin, keep `thesis`/`tension_statement` conservative and let downstream evidence skills strengthen the subsection.
- If contrast clusters collapse into overlapping paper pools, reroute before writing: after removing bridge papers, each side should still retain at least 2 unique papers.
- Use `bridge_terms` to surface concrete lexical handles that later evidence/ranking stages can still match (`OOD`, `sim-to-real`, `world model`, `failure detector`, specific benchmark families), not only generic axis names.
- Prefer domain-pack `cluster_rules` over ad-hoc bootstrap overlaps when the mapped set is already large enough to support disjoint clusters.
- Do not “fix” thin evidence by inventing more specific axes or stronger claims.


## Execution notes

When running in compatibility mode, `scripts/run.py` currently reads:
- `outline/outline.yml` for section/subsection structure
- `outline/mapping.tsv` for paper-to-subsection coverage
- `papers/paper_notes.jsonl` for structured evidence
- `GOAL.md` for topic/domain cues
- `outline/claim_evidence_matrix.md` as optional supporting context when present

## Script

### Quick Start

- `uv run python .codex/skills/subsection-briefs/scripts/run.py --workspace <workspace>`

### All Options

- `--workspace <dir>`
- `--unit-id <id>`
- `--inputs <a;b;...>`
- `--outputs <a;b;...>`
- `--checkpoint <C*>`

### Examples

- `uv run python .codex/skills/subsection-briefs/scripts/run.py --workspace <workspace>`

## Troubleshooting

- If the wrong domain pack is selected, inspect `GOAL.md` and the asset packs before changing the script.
- If briefs sound too generic, adjust the phrase/domain packs instead of adding more Python prose.
- If `papers/paper_notes.jsonl` is thin, reroute to note extraction rather than inventing axes.

Excerpt, first 5,020 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

1. 2026-10-19
   ✓SAFE B 89 / 100

   first audit · source 81994516a430

Questions about Subsection Briefs

• github.com/WILLOSCAR/research-units-pipeline-skills/tree/main/.c — the audited source at commit 81994516a430 and every finding aboveObserved · 19 Oct 2026

Spotted something wrong? Report a correction · Methodology