skvlt
AGENTS.md7 starsChanged 35 days agoMITView on GitHub
This repository is a lightweight Agent Skills manifest repo. The primary artifact is skvlt.yaml, which tracks approved skill sources, installed skill names, and manifest counts. The main companion document is MANIFEST_POLICY.md, which defines how agents should decide whether to add, replace, reject, or escalate manifest changes. There is currently no application code or automated test suite checked into this repository, but there is now a lightweight bun-based formatting toolchain and GitHub Actions formatting workflows. When editing skvlt.yaml: Follow MANIFEST_POLICY.md before…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository2 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
22 rules. About 0 are common, 14 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈14
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 25
Treat skvlt.yaml as the primary file.
- Very specific: line 26
Keep MANIFEST_POLICY.md aligned with actual maintenance behavior.
- Distinctive: line 27
Prefer small, auditable edits over broad manifest churn.
- Distinctive: line 28
Do not invent missing metadata about skills.
- Distinctive: line 29
Do not add agent-specific, repo-specific, or runtime-bound skills unless the policy explicitly allows them.
- Very specific: line 36
Keep total_sources equal to the number of source blocks.
- Very specific: line 37
Keep total_skills equal to the sum of all per-source counts.
- Very specific: line 38
Keep each source count equal to the number of listed skills in that block.
- Distinctive: line 39
Remove a source block entirely if its last skill is removed.
- Distinctive: line 40
Avoid duplicate skill names across the manifest.
- Distinctive: line 41
Prefer updating an existing trusted source block over adding a new source block.
- Distinctive: line 42
Treat source-level additions or removals as higher risk than skill-level additions or removals.
- Very specific: line 46
Follow MANIFEST_POLICY.md before making substantive manifest changes.
- Distinctive: line 50
Add specialized or dependency-closing skills only when they fit an already trusted source.
- Distinctive: line 51
When two general skills substantially overlap, prefer the one that is more complete, more practical, and more worth retaining.
- Distinctive: line 57
Before claiming a manifest edit is complete, verify:
- Distinctive: line 75
If you need upstream skill discovery, prefer checking the local installed skill files first, then use the relevant skills tooling or …
- Very specific: line 79
Update MANIFEST_POLICY.md when maintenance behavior changes.
- Distinctive: line 80
Update AGENTS.md when repository layout, validation steps, or working rules change.
- Distinctive: line 81
Keep docs concise and operational.
- Distinctive: line 85
Keep commits focused on one manifest or documentation change at a time.
- Very specific: line 88
Call out count updates explicitly if total_sources, total_skills, or source count fields changed.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Project Overview
This repository is a lightweight Agent Skills manifest repo.
The primary artifact is skvlt.yaml, which tracks approved skill sources, installed skill names, and manifest counts.
The main companion document is MANIFEST_POLICY.md, which defines how agents should decide whether to add, replace, reject, or escalate manifest changes.
There is currently no application code or automated test suite checked into this repository, but there is now a lightweight bun-based formatting toolchain and GitHub Actions formatting workflows.
Repository Layout
skvlt.yaml- source-of-truth manifest for installed skillsMANIFEST_POLICY.md- maintenance policy for agent-driven manifest changespackage.json- local formatter entrypoint and scripts.prettierrc.json- repository formatting rules.github/workflows/format-check.yml- formatting validation workflow.github/workflows/format-fix.yml- manual formatting remediation workflowLICENSE- repository license
Working Rules
- Treat
skvlt.yamlas the primary file. - Keep
MANIFEST_POLICY.mdaligned with actual maintenance behavior. - Prefer small, auditable edits over broad manifest churn.
- Do not invent missing metadata about skills; inspect local skill files or trusted upstream sources first.
- Do not add agent-specific, repo-specific, or runtime-bound skills unless the policy explicitly allows them.
Manifest Editing Instructions
When editing skvlt.yaml:
- Preserve the generated header comment if present.
- Keep
total_sourcesequal to the number of source blocks. - Keep
total_skillsequal to the sum of all per-source counts. - Keep each source
countequal to the number of listed skills in that block. - Remove a source block entirely if its last skill is removed.
- Avoid duplicate skill names across the manifest.
- Prefer updating an existing trusted source block over adding a new source block.
- Treat source-level additions or removals as higher risk than skill-level additions or removals.
Change Decision Rules
Follow MANIFEST_POLICY.md before making substantive manifest changes.
High-level defaults:
- Add specialized or dependency-closing skills only when they fit an already trusted source.
- When two general skills substantially overlap, prefer the one that is more complete, more practical, and more worth retaining; use maintenance and authority as tie-breakers rather than the primary decision rule.
- Reject skills that are tightly bound to a specific agent, repository, or runtime unless explicitly approved.
- Escalate ambiguous overlap, close calls between broad general skills, process-shaping changes, or source-level churn to a human.
Validation
Before claiming a manifest edit is complete, verify:
- repository formatting still passes
- the YAML remains readable
- the manifest counts are internally consistent
- the edited source block count matches the listed skills
- any dependency-based addition is actually justified by installed skills or policy
Useful local checks:
bun installbun run format:checkIf you need upstream skill discovery, prefer checking the local installed skill files first, then use the relevant skills tooling or trusted upstream listing.
Documentation Updates
- Update
MANIFEST_POLICY.mdwhen maintenance behavior changes. - Update
AGENTS.mdwhen repository layout, validation steps, or working rules change. - Keep docs concise and operational; this repo does not need broad product documentation.
Commit and Review Guidance
- Keep commits focused on one manifest or documentation change at a time.
- Mention the affected source or skill family in the commit message when possible.
- Summarize why a skill was added, replaced, rejected, or escalated.
- Call out count updates explicitly if
total_sources,total_skills, or sourcecountfields changed.
Common Gotchas
- Forgetting to update manifest counts after editing a source block
- Leaving an empty source block behind after removing its last skill
- Adding a skill dependency without adding the dependency itself
- Treating a specialized skill as a duplicate of a general skill
- Keeping a skill only because it exists locally, without checking whether it belongs in the manifest
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

