agentsclimarketplace

Release governance

Skill yha9806/academic-writing-toolkit/.claude/skills/release-governance

Local-first, evidence-controlled academic writing workflows for AI agents, with bounded revision, clean-room review, and release governance.

Install
npx -y skills add yha9806/academic-writing-toolkit --skill release-governance

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

What its author says it does

Copied from the file, not written here

Use when preparing, auditing, releasing, or rebuttal-hardening academic manuscripts, datasets, artifacts, reviewer packets, or claim registers involving multiple refs, local assets, human labels, or agent-assisted drafts.

SKILL.md

4.2 KB, 844 tokens by cl100k_base, as published. Nobody here has run it

/release-governance - Release Evidence Governance

Purpose

Prepare release-facing academic artifacts with explicit evidence control. The core rule is:

release truth = ref + artifact + gate

Do not infer release truth from memory, a branch name, a pull request title, a local file cache, or an agent draft.

Trigger Words

This skill activates on: release governance, release evidence, camera-ready, rebuttal packet, artifact packet, claim ledger, human evidence gate, release packet, /release-governance.

Core Rules

  1. Name the exact ref, artifact, and gate behind every release-facing claim.
  2. Keep draft advisory evidence, verified artifacts, and final human evidence separate.
  3. Do not promote agent or Gemini review output into final evidence.
  4. Do not treat ignored, untracked, cached, or local-only assets as submitted artifacts.
  5. Do not treat a repository check as venue, submission-system, or scientific compliance.
  6. If two refs diverge, name both and scope which one is canonical for each artifact.
  7. If a worktree is dirty, list the dirty paths and mark the packet as draft until changes are committed or explicitly scoped.

Evidence States

Use references/evidence_state_schema.md for the shared states:

  • draft_advisory
  • verified_artifact
  • human_final

human_final requires an explicit human confirmation gate when the project schema provides one. Agent drafts, reviewer suggestions, generated summaries, simulations, and prefilled labels stay draft_advisory.

Workflow

1. Declare Scope

Create release/release_scope.md with scope date, artifact name, intended use, included refs, excluded refs, and the highest-risk open question.

2. Map Repository Truth

Run narrow ref checks before reading a checkout as final state:

git fetch --all --prune
git status --short --branch
git worktree list --porcelain
git branch --all --verbose --no-abbrev

Create release/canonical_refs.csv. Detached HEADs are acceptable only when the exact SHA is recorded.

3. Inventory Assets

Create release/local_asset_inventory.csv to separate tracked, ignored, untracked, external-store, and generated assets. Local assets may support review, but release claims require an artifact anchor or explicit exclusion.

4. Anchor Artifacts And Claims

Create:

  • release/artifact_anchors.csv
  • release/claim_ledger.csv
  • release/evidence_gates.csv

Every paper-facing number, qualitative conclusion, table value, figure, dataset count, and human-label claim should point to the artifact and gate that support it.

5. Review Advisory Evidence

Gemini or another agent may review the scope, packet, or diff, but its result is advisory. Record advisory review in the verification report without changing any evidence state to human_final.

6. Verify Packet

Use references/release_workflow_templates.md for columns and report structure. If Python is available, run:

python {skill_dir}/scripts/check_release_packet.py <project_root>

The helper checks required packet files, CSV/JSON/YAML readability, evidence-state values, required columns, local absolute paths, and obvious template markers. YAML files use PyYAML when installed; otherwise the helper applies a basic local syntax check for simple mappings and lists. It does not run experiments, access networks, push branches, or judge scientific validity.

Stop Conditions

Stop and report a blocker if:

  • a final claim depends on a dirty or ambiguous ref
  • a human-label claim lacks a human confirmation gate
  • a release artifact exists only as an ignored, untracked, cache, or local-only file
  • a count comes from a pointer, stub, generated preview, or non-canonical ref
  • a draft pull request, closed-unmerged pull request, or advisory review is being treated as final release state
  • the packet validator reports issues that affect release-facing claims

What ships with it: 3 files

14.4 KB alongside SKILL.md, 1 of them executable

scripts/

Gives 0 of the 12 instructions most audit compliance skills give in 844 tokens

Counted across 937 of the 1,487 authors here whose files we hold, read 2026-08-07

  • Fetch latest guidelines before each reviewin 43 of 937, across 3 files
  • Group findings by severityin 43 of 937
  • Check files against all fetched rulesin 42 of 937, across 2 files
  • Output findings in terse file:line formatin 41 of 937, across 3 files
  • Ask user which files to review if none specifiedin 41 of 937, across 3 files
  • Read specified files or prompt user for filesin 39 of 937, across 1 file
  • Generate the audit reportin 33 of 937, across 30 files
  • Assign a severity to every findingin 25 of 937
  • Run automated accessibility scansin 23 of 937, across 13 files
  • Output a markdown audit reportin 22 of 937
  • Map findings to WCAG criteriain 20 of 937, across 10 files
  • Confirm audit scopein 19 of 937, across 9 files

Said here and by no other author read

  • name exact ref artifact and gate for each claim
  • separate draft advisory verified and human evidence
  • create release scope document
  • map repository truth using git commands
  • record canonical refs in a CSV file
  • inventory local assets in a CSV file

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 327,069. 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.