Add AGENTS.md

[masenf]

Get some sanity when developing with Claude, Cursor, and friends.

[greptile-apps]

Greptile Summary

Introduces AGENTS.md as the canonical coding-agent guidelines file and makes CLAUDE.md a symlink to it, so tools that look for either filename get identical content from a single source. CONTRIBUTING.md is simultaneously cleaned up by removing the darglint documentation-check step.

• The coverage comment in AGENTS.md (>=72%) contradicts the still-unchanged CONTRIBUTING.md text ("below 70%"), which could confuse human contributors or cause surprise CI failures.

Confidence Score: 5/5

Safe to merge; only documentation files changed, with one minor coverage-threshold inconsistency between AGENTS.md and CONTRIBUTING.md.

All three changed files are documentation/config with no runtime impact. The single finding is a P2 documentation inconsistency (72% vs 70% coverage floor) that doesn't block functionality.

AGENTS.md — coverage threshold comment should be reconciled with CONTRIBUTING.md

Important Files Changed

Filename	Overview
AGENTS.md	New AI agent guidelines file; minor coverage threshold (72%) contradicts CONTRIBUTING.md (70%)
CLAUDE.md	New symlink pointing to AGENTS.md so Claude Code and other tools reading CLAUDE.md get the same content
CONTRIBUTING.md	Removes darglint from required pre-commit checks and contributing instructions; clean and consistent removal

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Claude Code / Claude CLI] -->|reads CLAUDE.md| S[CLAUDE.md symlink]
    B[OpenAI Codex / Cursor] -->|reads AGENTS.md| C[AGENTS.md source file]
    C2[Other tools e.g. Gemini] -->|reads AGENTS.md| C
    S -->|resolves to| C
    C --> G[Single set of coding guidelines]

Loading

_{Reviews (1): Last reviewed commit: "Add AGENTS.md" | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 9 untouched benchmarks

---

_{Comparing masenf/agents-md (434c2b6) with main (4e7b239)}

[FarhanAliRaza]

I think some things can be added here; I don't know if they are just my personal preference or can be added globally. But these things are what i noticed AI does badly generally.

About performance

• Do not add more work-intensive functions like (isinstance checks) to bypass naive problems (like type checking). Fix code at the fundamental level.
• Do not repeat validation and be over-defensive. (by adding if conditions everywhere to revalidate already validated data upstream)
• Think about performance in terms of machine-level CPU cycles, like data copy is always slow.

About Testing

• We do not want bogus tests that pass even the thing it is testing is failing.
• We want to use RED and GREEN test-driven development..
• Tests should fail without bug fixes that we did.

[masenf]

We want to use RED and GREEN test-driven development..
Tests should fail without bug fixes that we did.

I think these two are already represented in the current doc. I'll add the others