agentskills.codes
WM

wm-gap-auditor

by RICHAAARCSource

Audit implementation gaps and risks between the current watermarking project and stated research goals. Use when users ask for gap audit, risk audit, goal alignment, static code audit, or a split report of static-confirmable gaps vs gaps that require real SD workflow.

Install

mkdir -p .claude/skills/wm-gap-auditor && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15286" && unzip -o skill.zip -d .claude/skills/wm-gap-auditor && rm skill.zip

Installs to .claude/skills/wm-gap-auditor

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Audit implementation gaps and risks between the current watermarking project and stated research goals. Use when users ask for gap audit, risk audit, goal alignment, static code audit, or a split report of static-confirmable gaps vs gaps that require real SD workflow.
268 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)

About this skill

wm-gap-auditor

Produce a defensible gap audit tied to repository evidence.

Execute

  1. Extract target claims from user text and project docs.
  2. Build a claim-to-anchor map with configs/ and main/ as the core release truth source; use scripts/ and tests/ only as auxiliary evidence.
  3. Validate each claim with static evidence first:
  • Read implementation paths.
  • Read contract/config paths.
  • Run only relevant tests or audit scripts when they can decide the claim, but do not let helper script behavior override main/.
  1. Classify each gap item into exactly one bucket:
  • static-audit-confirmable
  • requires-real-sd-workflow
  1. If the user mandates a fixed phrase for undecidable items, output that phrase exactly.

Required Output Shape

For every item include:

  • gap_item
  • risk_level (high/medium/low)
  • evidence (file path or test name)
  • reasoning
  • fix_or_validation_action

Repository Anchors

Prioritize these anchors before expanding search:

  • configs/paper_faithfulness_spec.yaml
  • notebook/Colab_Workflow.ipynb
  • main/watermarking/content_chain/
  • main/watermarking/geometry_chain/
  • main/watermarking/fusion/
  • main/policy/
  • tests/test_onefile_workflow.py
  • tests/test_publish_workflow_uses_paper_profile_by_default.py
  • tests/test_paper_faithfulness_gate.py

Use these only as auxiliary workflow anchors after core files:

  • scripts/run_onefile_workflow.py
  • notebook/Colab_Workflow.ipynb

Classification Rules

  • Put into static bucket if code/contracts/tests already prove presence, absence, or violation.
  • Put into real-workflow bucket if conclusion depends on SD model runtime behavior, generated image quality, attack robustness curves, or end-to-end geometry recovery under real inference.

Search skills

Search the agent skills registry