pi-skills / config
AGENTS.md3 starsChanged 2 months agoMITView on GitHub
Before finalizing non-trivial implementation:
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 repository17 other files it gives its agents.
- analyzeSkill
- app-resizabilitySkill
- code-reviewSkill
- device-interactionSkill
- figmaSkill
- git-commitSkill
- gitlab-glabSkill
- git-redateSkill
- jira-task-contextSkill
- jira-updateSkill
- mr-comment-resolutionSkill
- pragmatic-testingSkill
- resolve-merge-conflictsSkill
- swift-concurrencySkill
- tddSkill
- verifySkill
- xcodebuild-xcsiftSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 5
Treat this section as the source of truth for mutation authorization.
- Distinctive: line 8
If implementation was not explicitly requested, present a concise approach, risks, and verification plan, then ask for approval.
- Distinctive: line 10
Do not stage or commit changes unless the user directly requests it.
- Very specific: line 21
Before adding, changing, reviewing, or deleting tests or test doubles, use pragmatic-testing, whether working test-first or through normal …
- Very specific: line 26
When invoking shell xcodebuild, use xcodebuild-xcsift and pipe output through xcsift.
- Distinctive: line 31
Before finalizing non-trivial implementation:
- Distinctive: line 36
Do not treat a successful build as proof that tests or product behavior are correct.
≈ “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
Global Pi Workflow
Change safety and authorization
- Treat this section as the source of truth for mutation authorization; skills may add scope or evidence requirements but must not require a redundant confirmation.
- Inspect relevant code, tests, project guidance, and repository state before making non-trivial changes.
- For investigation, explanation, planning, review, or context lookup, remain read-only.
- If implementation was not explicitly requested, present a concise approach, risks, and verification plan, then ask for approval.
- An explicit request to implement, fix, update, or make a change authorizes the scoped local edits and ordinary verification after inspection. Do not ask for a second confirmation solely because a skill was loaded.
- Do not stage or commit changes unless the user directly requests it.
- Preserve unrelated user changes.
- Destructive Git or filesystem operations require an explicit request for that action. Preview the exact effect when practical, and never stash changes automatically.
- Externally visible mutations require explicit authorization for the specific action. A direct request in the same message counts; otherwise present the exact action and ask before executing it.
Workflow routing
- For Jira-backed implementation tasks, use
analyze; for Jira-only context lookup, usejira-task-context. - For code reviews, branch reviews, local-change reviews, and GitLab MR reviews, use
code-reviewand remain read-only. - For GitLab access, use
gitlab-glaband the authenticatedglabCLI. - For MR reviewer-comment work, use
mr-comment-resolution. - Before adding, changing, reviewing, or deleting tests or test doubles, use
pragmatic-testing, whether working test-first or through normal implementation. During review or investigation, apply it read-only. - For commits, use
git-commit. - For Figma links, design inspection, implementation from designs, and explicitly requested Figma edits, use
figmawith the project-enabled native Figma MCP. Local implementation does not authorize Figma writes. - For Xcode-specific builds, tests, diagnostics, SwiftUI previews, Apple documentation, and Swift snippets, prefer the available Xcode MCP tools.
- Fall back to shell
xcodebuildonly when Xcode MCP is unavailable or does not expose the required operation. A real build or test failure reported by Xcode MCP is not a reason to rerun the same check through the shell, and the two routes must not run concurrently. - When invoking shell
xcodebuild, usexcodebuild-xcsiftand pipe output throughxcsift. - Access GitHub resources through authenticated
gh, not unauthenticated web fetching. If access fails, checkgh auth statusrather than falling back to public web access.
Verification
Before finalizing non-trivial implementation:
- Re-read the request and any Jira, specification, or design context used.
- Inspect the final diff against the requested behavior and acceptance criteria.
- Identify and run focused tests, or state exactly why they were not run.
- Do not treat a successful build as proof that tests or product behavior are correct.
- For visual SwiftUI changes, render and inspect representative previews when Xcode MCP is available. Compare the rendered snapshot with the supplied screenshot, Figma design, or requested state before claiming visual correctness.
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.

