Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write the survey's front matter files (Abstract, Introduction, Related Work, Discussion, Conclusion) in paper voice, with high citation density and a single evidence-policy paragraph. **Trigger**: front matter writer, introduction writer, related work writer, abstract writer, discussion writer, conclusion writer, 引言, 相关工作, 摘要, 讨论, 结论.
.claude/skills/willoscar-front-matter-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -6% | 0% |
Purpose: produce the paper-level shell while preserving the current output contract and approval gate.
Always read:
references/overview.mdRead by task:
references/abstract_archetypes.md for sections/abstract.mdreferences/context_projection.md when adjusting how chapter briefs are projected into Introduction / Related Work contextreferences/introduction_jobs.md for Introduction body filesreferences/related_work_positioning.md for Related Work body filesreferences/discussion_conclusion_patterns.md for sections/discussion.md and sections/conclusion.mdreferences/forbidden_stems.md before revising any reader-facing prosereferences/examples_good.md and references/examples_bad.md for calibration onlyMachine-readable assets:
assets/front_matter_context.schema.jsonassets/front_matter_context_schema.json (compatibility alias for checklist-aligned naming)assets/front_matter_contract.jsonassets/front_matter_templates.jsonassets/front_matter_context_projection.jsonRequired:
DECISIONS.md with Approve C2outline/outline.ymloutline/mapping.tsvcitations/ref.bibOptional but useful:
GOAL.mdqueries.mdpapers/retrieval_report.mdpapers/core_set.csvoutline/coverage_report.mdoutline/writer_context_packs.jsonlKeep the current output contract:
sections/abstract.mdsections/S<sec_id>.md for front-matter H2 bodiessections/discussion.mdsections/conclusion.mdoutput/FRONT_MATTER_REPORT.mdoutput/FRONT_MATTER_CONTEXT.json, including the base template bank, anyselected domain overlay, and their repository-relative SHA-256 hashes
Current mode is reference-first with script compatibility:
references/assets/front_matter_contract.jsonassets/front_matter_templates.jsonscripts/run.py still owns approval checks, metadata parsing, deterministic hook selection, file emission, and context-sidecar writingThat means:
SKILL.mdUse scripts/run.py as a deterministic helper for:
assets/front_matter_contract.json + assets/front_matter_templates.jsonDo not treat the script as the main place for long-form writing guidance.
Keep these stable:
sections/abstract.md starts with ## Abstract or ## 摘要sections/S<sec_id>.md files are body-onlysections/discussion.md includes ## Discussionsections/conclusion.md includes ## Conclusionuv run python .codex/skills/front-matter-writer/scripts/run.py --workspace <workspace>references/forbidden_stems.md and references/examples_bad.md.references/related_work_positioning.md.SKILL.md again.When running this skill in compatibility mode, scripts/run.py currently reads these inputs directly:
DECISIONS.md for Approve C2outline/outline.yml to resolve Introduction / Related Work section idsoutline/mapping.tsv for front-matter citation placement contextcitations/ref.bib for citation-key scopeGOAL.md, queries.md, papers/retrieval_report.md, papers/core_set.csv, outline/coverage_report.md, and outline/writer_context_packs.jsonl as optional context sourcesuv run python .codex/skills/front-matter-writer/scripts/run.py --workspace <workspace>--workspace <dir>--unit-id <id>--inputs <path1;path2>--outputs <path1;path2>--checkpoint <C*>uv run python .codex/skills/front-matter-writer/scripts/run.py --workspace <workspace>DECISIONS.md lacks Approve C2, the script will write a checkpoint block and exit.citations/ref.bib or outline/outline.yml is missing, restore those inputs before rerunning.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-14 | fail→pass | 9,944 | 2,142 | -78% | 1 | 1 | 0% | 1,427 | 1,614 | +13% | 0 | 0 | — |
case-01 | fail→fail | 25,231 | 5,975 | -76% | 1 | 1 | 0% | 3,721 | 1,657 | -55% | 0 | 0 | — |
case-02 | fail→fail | 27,794 | 4,980 | -82% | 1 | 1 | 0% | 4,417 | 1,546 | -65% | 0 | 0 | — |
case-03 | fail→fail | 33,801 | 5,554 | -84% | 1 | 1 | 0% | 5,641 | 1,533 | -73% | 0 | 0 | — |
case-04 | fail→pass | 8,655 | 3,454 | -60% | 1 | 1 | 0% | 1,238 | 1,853 | +50% | 0 | 0 | — |
case-05 | fail→pass | 7,803 | 2,919 | -63% | 1 | 1 | 0% | 1,239 | 1,773 | +43% | 0 | 0 | — |
case-06 | fail→fail | 11,443 | 2,226 | -81% | 1 | 1 | 0% | 1,796 | 1,651 | -8% | 0 | 0 | — |
case-07 | pass→fail | 11,115 | 2,001 | -82% | 1 | 1 | 0% | 1,695 | 1,595 | -6% | 0 | 0 | — |
case-08 | pass→pass | 6,520 | 2,445 | -63% | 1 | 1 | 0% | 941 | 1,681 | +79% | 0 | 0 | — |
case-09 | fail→pass | 7,726 | 2,428 | -69% | 1 | 1 | 0% | 1,204 | 1,703 | +41% | 0 | 0 | — |
case-10 | fail→pass | 12,840 | 4,039 | -69% | 1 | 1 | 0% | 1,932 | 1,812 | -6% | 0 | 0 | — |
case-11 | fail→pass | 9,103 | 2,715 | -70% | 1 | 1 | 0% | 1,436 | 1,725 | +20% | 0 | 0 | — |
case-12 | fail→fail | 12,779 | 7,094 | -44% | 1 | 1 | 0% | 2,093 | 2,490 | +19% | 0 | 0 | — |
case-13 | fail→pass | 13,957 | 4,461 | -68% | 1 | 1 | 0% | 2,260 | 2,019 | -11% | 0 | 0 | — |
case-15 | fail→pass | 7,828 | 1,828 | -77% | 1 | 1 | 0% | 1,196 | 1,599 | +34% | 0 | 0 | — |
case-16 | fail→pass | 13,444 | 6,812 | -49% | 1 | 1 | 0% | 2,042 | 2,365 | +16% | 0 | 0 | — |
case-17 | fail→pass | 14,105 | 1,833 | -87% | 1 | 1 | 0% | 2,106 | 1,574 | -25% | 0 | 0 | — |
case-18 | fail→pass | 8,374 | 3,010 | -64% | 1 | 1 | 0% | 1,252 | 1,777 | +42% | 0 | 0 | — |
case-19 | fail→pass | 9,434 | 2,138 | -77% | 1 | 1 | 0% | 1,411 | 1,601 | +13% | 0 | 0 | — |
case-20 | pass→pass | 13,621 | 14,644 | +8% | 1 | 1 | 0% | 2,224 | 3,629 | +63% | 0 | 0 | — |
case-21 | pass→pass | 12,064 | 9,582 | -21% | 1 | 1 | 0% | 1,874 | 2,883 | +54% | 0 | 0 | — |
case-22 | pass→pass | 6,755 | 5,479 | -19% | 1 | 1 | 0% | 1,175 | 2,182 | +86% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.