Publisher: Sentry · Editorial profile by AI Tools Ratings · Review method
What task is this skill designed for?
Build or repair an AGENTS.md file that helps a coding agent find the right commands and conventions. Best when project documentation has grown inconsistent and agents repeatedly use obsolete instructions.
What input and output should you expect?
Input: Provide repository layout, package and lock files, CI configuration, existing root and nested instructions, and the repeated mistakes the new guidance should prevent.
Output: A compact instruction file or proposed patch, with scoped links to longer documentation. The source prefers short, verified guidance over duplicating the entire README.
What does it need to run?
Repository read access and, for applying the result, permission to edit instruction files. Verify command definitions against the checked-out project.
Publisher documents Claude Code installation and skills.sh support for compatible agents including Cursor, Cline and GitHub Copilot. Host permissions and invocation behavior still require validation.
How should you approach installation?
- Read the pinned SKILL.md and repository LICENSE; compare project-specific conventions with your own instructions.
- Follow the publisher installation guidance for your host, retaining the full skill folder and references; begin in a disposable repository with minimal permissions.
- Run the suggested draft or read-only trial, review its acceptance checklist, and grant write or remote permissions only for the intended task.
Installation guide: discovery, dependencies and runtime checks
What files, accounts or data can it access?
Can write AGENTS.md and related instruction links; these files influence future agent behavior. Repository context may be sent to the host model. Do not copy credentials or unsafe commands from untrusted documents.
What does the skill cost?
No separate Skill purchase price is stated in the reviewed repository. The host agent subscription or model usage is separate; repository services and any authorized CI jobs may add costs. A local drafting trial does not require paid remote execution.
What are its limits and alternatives?
Instruction hierarchy and symlink support vary by host and operating system. The source includes Sentry attribution conventions that should not be imposed on unrelated repositories. A plausible command still needs validation.
Use a README for human onboarding or narrowly scoped nested instructions for rules that apply only to one package.
What could you ask in a small trial?
Try this with a small sample:
Propose a short AGENTS.md for this sample repository. The README says npm test but package.json defines test:unit; identify the discrepancy using files, do not guess a command. Preserve existing user constraints and do not create symlinks or edit files yet.
How would you verify the result?
- Every named path and command exists in the current repository.
- General guidance is separated from rules limited to a package.
- The draft contains no secrets, copied project-irrelevant conventions or unsupported validation claims.
Use a baseline and record acceptance evidence before treating a successful task as a reliable workflow.
Which source and license were checked?
Read the reviewed SKILL.md at its pinned commit · Complete reviewed folder
Repository and current publisher guidance
Checked:
Apache-2.0 — repository LICENSE at the reviewed commit. License evidence
Version details and review scope
Source: getsentry/skills/skills/agents-md
Reviewed commit: d12f250460ffd961f549dbebf7eeb0cf4afd4c2e
SKILL.md SHA-256: 8a91ada35068f43ef7ee03e38937863bac4f407878b06f0a4f1ddb4af24bec7e
Source-document review. This profile covers the source workflow, stated dependencies and license evidence. See our editorial methodology for how we distinguish source reviews and hands-on evaluations.