agentsclimarketplace
Sign in

Self review

Skill yha9806/academic-writing-toolkit/.claude/skills/self-review

Review the user's own manuscript, paper, thesis chapter, rebuttal, or release packet with clean-room anti-contamination controls. Use when asked for self-review, internal review, pre-submission review, readiness check, reviewer simulation on own work, or claim-evidence self-audit where prior chat memory, unstated assumptions, model background knowledge, or unlisted local notes must not be treated as evidence.From its SKILL.md

Install
npx -y skills add yha9806/academic-writing-toolkit --skill self-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.9 KB, 952 tokens by cl100k_base, as published. Nobody here has run it

/self-review - Clean-Room Manuscript Self-Review

Purpose

Audit the user's own work without letting memory, prior chats, unstated project knowledge, or the model's background knowledge become evidence.

The governing rule is:

self-review truth = explicit review packet + source anchor

Use /argument-governance first when the manuscript needs a formal intent, contribution, claim, and evidence map.

Codex-Only Baseline

Complete self-review with Codex, the review manifest, allowed sources, and the bundled packet checker. Do not require Gemini, gemini-agent, a second model, or a subagent. If an external review is available, keep it in Reviewer-risk inference or advisory notes and never use it as source support.

If /argument-governance is unavailable, manually extract the same clean-room argument spine from manifest-listed sources only.

Enhanced Advisory Mode

If the manifest and the user explicitly allow an API-key-backed advisory review, Codex may run or incorporate a second-model pass after the clean-room self-review packet is valid.

Rules:

  • the base clean-room review must be possible without the external call
  • API keys must be read from environment variables only
  • the manifest may name api_key_env_var, but must never store the key value
  • only manifest-approved source subsets may be sent externally
  • external findings must be placed under Reviewer-risk inference or advisory notes
  • external findings must be re-grounded against allowed sources before becoming revision actions
  • unsupported external comments stay unsupported

Core Rules

  1. Use only files listed in the review manifest.
  2. Treat prior chat memory, unstated project assumptions, model background knowledge, and unlisted notes as forbidden evidence.
  3. Split every finding into Supported by packet, Not supported by packet, or Reviewer-risk inference.
  4. Every supported finding must include a source anchor.
  5. Do not repair missing evidence by remembering earlier conversations.
  6. Do not treat generated reviews, agent drafts, or reviewer simulations as final evidence.
  7. Do not edit the manuscript until the user approves specific revision actions.
  8. Do not treat an unavailable external review tool as a blocker.

Required Packet

The preferred layout is:

review_packet/
  review_manifest.yaml
  manuscript.md or manuscript.pdf
  references.bib
  evidence/
  figures/
  tables/
  claims/

Read references/clean_room_protocol.md before reviewing. Read references/self_review_packet_schema.md before creating or validating a packet.

Workflow

1. Validate The Clean-Room Packet

Resolve the bundled helper at scripts/check_self_review_packet.py relative to this SKILL.md, then run:

python3 {skill_dir}/scripts/check_self_review_packet.py review_packet --json

If the packet is missing or invalid, report the issue before reviewing.

2. Build The Source-Bounded Reading List

Read only manifest-listed files. If a needed file is not listed, ask whether to add it to the manifest or mark the issue as unsupported.

3. Extract The Argument Spine

From the packet only, extract:

  • stated intent
  • named gap
  • contributions
  • main claims
  • evidence anchors
  • limitations
  • reviewer-risk areas

4. Run Self-Review Checks

Check:

  • gap-contribution alignment
  • claim hierarchy
  • claim-evidence fit
  • evidence balance
  • unsupported or overextended claims
  • missing limitations
  • reviewer attacks with weak defenses
  • internal consistency and submission blockers

5. Write A Clean-Room Report

The report must separate:

  • Supported by packet
  • Not supported by packet
  • Reviewer-risk inference

Do not merge these categories.

Output Pattern

## Clean-Room Self-Review

### Packet Boundary
- Manifest:
- Allowed sources:
- Forbidden sources:

### Supported By Packet
| Finding | Source anchor | Severity | Action |

### Not Supported By Packet
| Claim or need | Missing source | Risk | Action |

### Reviewer-Risk Inference
| Risk | Basis in packet | Why it matters | Action |

### Revision Actions
| Priority | Action | Requires new evidence? |

Stop Conditions

Stop and report a blocker if:

  • no review manifest is provided
  • the user asks to rely on previous chat or memory for evidence
  • a central claim needs a source not listed in the manifest
  • the packet validator reports missing allowed files
  • the user asks to mark unsupported claims as supported

What ships with it: 3 files

10.9 KB alongside SKILL.md, 1 of them executable

references/

scripts/

Gives 0 of the 12 instructions most review quality skills give in 952 tokens

Counted across 1,048 of the 1,783 authors here whose files we hold, read 2026-08-07

  • Ask questions one at a timein 81 of 1048, across 64 files
  • Provide a recommended answer for each questionin 73 of 1048, across 50 files
  • Explore the codebase instead of asking answerable questionsin 66 of 1048, across 42 files
  • Resolve dependencies between decisions one-by-onein 42 of 1048, across 17 files
  • Interview the user relentlessly about the planin 38 of 1048, across 13 files
  • Order findings by severityin 31 of 1048
  • Resolve each branch of the decision treein 27 of 1048, across 5 files
  • Run a grilling sessionin 26 of 1048, across 5 files
  • Update CONTEXT.md immediately when a term is resolvedin 26 of 1048, across 11 files
  • Propose precise canonical terms for vague languagein 25 of 1048, across 7 files
  • Create documentation files lazilyin 24 of 1048, across 5 files
  • Assign severity to every findingin 24 of 1048

Said here and by no other author read

  • use only files listed in the review manifest
  • split every finding into three support categories
  • include a source anchor for every supported finding
  • validate the review packet before reviewing
  • read only manifest-listed files
  • extract the argument spine from the packet

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,790. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.