- Home
- Skills
- Prompt & Skill Engineering
- skill conflict detector
Works with the AI Agents you already use
skill conflict detector
Audit your entire Claude Code setup for behavioral contradictions, precedence gaps, and stale instructions.
$12
skill conflict detector
Example session with this skill installed
I have 30 skills installed and just updated my CLAUDE.md. Audit my whole setup for behavioral contradictions or stale instructions.
- Read your context and instructions
- Compiled the conflict detector
I have audited 32 skills across .claude/skills/, 2 CLAUDE.md files, and 3 active hooks. I found 3 behavioral conflicts where instructions overlap or contradict current policy.
Conflict Audit Report
| Class | Source A | Source B | Conflict Summary |
|---|---|---|---|
| (a) Direct | skill-git-pro | CLAUDE.md | Skill requires npm test before every commit; CLAUDE.md rule #4 allows [skip-test] flag. |
| (b) Ambiguity | react-refactor | typescript-helper | Both claim the refactor component trigger with no stated precedence. |
| (c) Stale | deploy-bot | Project Policy | Skill references a jenkins-ci hook that was replaced by GitHub Actions in June. |
Findings Detail
Finding 1: Direct Contradiction (Class A)
- Trigger: Any
git commitcommand. - Skill A (
skill-git-pro):"Always run 'npm test' and ensure a passing build before attempting a commit." - Source B (
CLAUDE.md):"Commits may bypass local testing if the message contains the [skip-test] string." - Result: If a user includes the flag, the agent receives two incompatible instructions for the same action.
Finding 2: Precedence Ambiguity (Class B)
- Trigger: Requests matching
refactor * component. - Skill A (
react-refactor): Handles component logic cleanup. - Skill B (
typescript-helper): Handles type definition optimization. - Result: No instruction exists to determine if these should run sequentially or if one takes priority.
Finding 3: Stale Contradiction (Class C)
- Trigger: Deployment workflows.
- Source:
deploy-botinstructions quote:"Refer to CLAUDE.md section 'CI-GATES' for auth keys."
Observation: The 'CI-GATES' section was removed from CLAUDE.md on 2024-11-12. The skill is referencing orphaned data.
Next steps
- Add a precedence rule to
CLAUDE.mdfor refactoring tasks. - Update
skill-git-proto recognize the[skip-test]override. - Remove the stale Jenkins reference from
deploy-bot.
Connects securely to your tools. The creator never sees your data.
About this skill
As a Claude Code library grows, contradictions become inevitable. You append a rule to CLAUDE.md, install a new marketplace skill, and wire in a guard hook, but nothing checks if these pieces actually agree. One skill might demand a build before every commit while another allows skipping it via a flag. When both apply to the same request, the agent is left with ambiguous instructions, leading to unpredictable behavior or silent failures.
What it does
- Multi-source auditing reads every
SKILL.md,CLAUDE.md, and hook script in your setup to find hidden behavioral conflicts. - Instruction extraction pulls imperatives and their trigger conditions to identify where two rules fight for the same input.
- Conflict classification categorizes issues into direct contradictions, precedence ambiguities, stale references, or hook-vs-instruction mismatches.
- Precedence mapping identifies triggers where no clear "winner" is defined, preventing the agent from guessing which rule to follow.
How it works
- Discovery phase enumerates all installed skills, local and global config files, and the actual logic inside your hook scripts.
- Behavioral bucketing groups instructions by tool name, file path, or lifecycle event to find overlaps at scale without noise.
- Conflict reporting quotes the exact conflicting sentences and identifies the shared trigger condition for every finding.
Frameworks & tools
Specifically built for Claude Code environments using .claude/skills/, CLAUDE.md configuration files, and settings.json hook definitions.
Why this beats prompting it yourself
Standard prompts only see the current context window or one file at a time. This skill performs a systematic pairwise comparison across your entire library, including reading the source code of your shell hooks to ensure your instructions match your actual enforcement.
Use cases
- Auditing a large organization-wide skill library for instructions that have drifted out of sync.
- Checking if a new
CLAUDE.mdpolicy contradicts existing skills written months ago. - Verifying that a new guard hook won't silently break a skill's core assumptions.
Known limitations
Does not optimize skill trigger descriptions or check for commercial duplication. Requires read access to hook scripts to perform deep logic analysis.
Does not optimize trigger descriptions or check for commercial duplication. Requires read access to hook scripts for logic analysis.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago