Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsSkill Inspector
Inspects any SKILL.md agent skill package against the open standard, grades every check with evidence, fixes what it finds, and writes a verified inspection report.
$10
Skill Inspector
Example session with this skill installed
Inspect my skill package at ./recipe-cost-calculator before I publish it. Fix mode.
- Read your context and instructions
- Compiled the inspector
- Generated the data export
Inspected recipe-cost-calculator in fix mode. The first pass found 4 Fails and 2 advisories, so the package was not publish-ready. All fixes are applied and re-verified, and the final verdict is Certified.
Inspection Report: recipe-cost-calculator
Date: 2026-10-21 Mode: fix | Files read: 2
Verdict (first pass)
Needs fixes. Four checks broke the standard: a missing description trigger, a reference to a file that is not in the package, a proprietary harness component name, and a credential-shaped literal shipped in the body.
Scorecard (first pass)
| Check | Result | Evidence |
|---|---|---|
| 1 Frontmatter parses | Pass | SKILL.md:1-3 |
| 2 Name field | Pass | SKILL.md:2 |
| 3 Description trigger | Fail | frontmatter holds name only, no description |
| 4 Body line budget | Pass | 18 lines |
| 5 References resolve | Fail | SKILL.md:16 prices.csv not in package; cost-report.md is a workflow output, exempt |
| 6 Support-file split | Pass | reference data lives in ingredients.md |
| 7 Orphan files | Pass | 2 files, both used |
| 8 Portability | Fail | SKILL.md:9 names a proprietary harness chart component |
| 9 Register quality | Advisory | SKILL.md:7 em dash character |
| 10 Scope match | Advisory | SKILL.md:18 claims network and terminal access that no workflow step uses |
| 11 Credential patterns | Fail | SKILL.md:11 key shape sk-live-9f8e**** |
| 12 Personal data | Pass | none found |
| 13 Encoded blobs | Pass | none found |
| 14 Undeclared access | Pass | no scripts in package |
Findings
- Check 3 Fail, SKILL.md frontmatter: no description field. The description is the trigger that tells the host agent when to load this skill. Without it, the skill only loads if the buyer types its exact name, which kills discovery on any platform. Fix path: write a two sentence third person trigger stating what the skill does and when to use it.
- Check 5 Fail, SKILL.md:16: the body tells the agent to look up prices in prices.csv, and the package holds no prices.csv. The skill breaks on first use. Fix path: ship the file, or reword the step to take prices from the user.
- Check 8 Fail, SKILL.md:9: the body names a proprietary harness chart component. The skill stops working the moment it runs anywhere else, which is the core portability rule of the standard. Fix path: reword to plain Markdown output.
- Check 11 Fail, SKILL.md:11: a live-shaped API key literal sits in the body. Even if the key is fake, credential shapes in skill files get flagged by platform reviewers and teach the buyer's agent to treat secrets as normal content. Fix path: remove the line and replace it with a rule against embedding credentials.
Fixes applied
| Fix | Check | Re-check |
|---|---|---|
| Added a two sentence description trigger to the frontmatter | 3 | Pass |
| Reworded the price step to read prices from the user or a user-provided file | 5 | Pass |
| Replaced the harness component line with plain Markdown output wording | 8 | Pass |
| Removed the credential literal, added a rule against embedding credentials | 11 | Pass |
| Replaced the em dash with a period | 9 | Pass |
| Removed the network and terminal access claim | 10 | Pass |
Verdict (after fixes)
Certified. All 14 checks pass on the re-run: the frontmatter carries a working description trigger, every reference resolves, the output wording is portable, and no credential shapes remain.
Next steps
- Publish as-is. The package now meets the open SKILL.md standard.
- Re-run this inspection after any edit, especially before version bumps.
- Rotate the exposed key if it was ever real. A credential shape in a shipped file is treated as burned.
prices.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Skill Inspector certifies agent skill packages before they ship. Point it at a folder holding a SKILL.md and it reads every file, runs a 14-point inspection against the open standard, grades each check Pass, Advisory, or Fail with file and line evidence, and writes a clean inspection report. In fix mode it repairs what it finds, re-runs the full check set, and records the before and after.
What it does
- Structure: frontmatter parse, name and description trigger quality, body line budget
- Disclosure: broken file references, orphan files, support-file split
- Wording: harness-specific references that break portability, register quality advisories
- Scope: declared permissions against actual contents, overreach and underclaim
- Safety: credential shapes, personal data, encoded blobs, undeclared network or environment access
How it works
- Point it at a skill folder and pick report-only or fix mode
- It reads every file in the package without installing or executing anything
- Every check in the bundled rubric is graded with file and line evidence
- Findings map to documented fix patterns; safety findings are masked and left to your decision
- It writes an inspection report you can hand to a reviewer, a buyer, or a team
Why this beats prompting it yourself
- A one-off prompt forgets the standard between sessions; the bundled rubric runs all 14 checks every time
- Findings cite file and line, so the report stands as evidence
- Fix mode re-runs the whole check set, so a fixed package is certified with the before and after on record
- The rubric separates rule violations from advisories, so you know what blocks shipping and what only weakens quality
Known limitations
- The bundled standard reference dates October 2026; verify your target platform's current publishing rules before shipping
- Static inspection only; the inspector reads files and does not judge runtime behavior
- Safety findings are reported with masked values and are left to your decision
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 passes our trust checks below.
Security scanned
Passed 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 scanned
Verified clean today
- Passed all security checks, Safe to install