Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/git-ai-project/git-ai/agents-mdgit clone --depth 1 https://github.com/git-ai-project/git-aiWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.04559 | $0.04559 |
| Opus 5 | $0.02279 | $0.02279 |
| Sonnet 5 | $0.00912 | $0.00912 |
| Haiku 4.5 | $0.00456 | $0.00456 |
Grade A, and why
git-ai AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 249 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Non-Negotiable Rules
These are hard constraints. Violating any of them will get a PR rejected outright.
-
All Git integration/processing is trace2-driven -- we do NOT wrap git. All git command processing is based on trace2, so EVERYTHING we do must be fully async. As a result, we cannot rely on repo/git state at processing time being reflective of the state whenever the given operation actually occurred. We have a highly latency-sensitive trace2 ingestion flow that gathers/estimates the minimal possible state and orders events for later async processing.
-
No new work on the critical ingestion path. You CANNOT add git spawns, git object lookups, ref checks, etc. in the critical ingestion path of the daemon. It is EXTREMELY latency-sensitive -- even additional file reads have meaningful overhead on this path, which is sensitive to sub-millisecond latency increases.
-
No non-constant-time git work anywhere. You CANNOT add git spawns, git object lookups, ref checks, etc. that are not constant-time (e.g., must be O(2)/O(3), not O(n)). You cannot have any operation that calls git per commit, per file, per object, or per ref -- that is not constant time. You can still call git outside of latency-sensitive paths, but instead of calling it once per item, find a constant-time alternative: a plumbing command, a command that accepts batched inputs, etc., so you avoid the massive overhead of N process spawns. Rule of thumb: if you're calling git or looking up objects/refs "for each
<thing>," you're doing it wrong. -
Unbounded git spawns are an automatic rejection. Changes/PRs with unbounded git spawns will NOT be accepted.
-
Reuse existing code. This is a large codebase. For nearly any operation you need, there is almost certainly a helper function already available and tested. Reuse code as much as practical -- it also keeps diffs small. PRs that fail to reuse existing code where applicable will be rejected immediately.
-
Follow STRICT TDD. PRs that are not clearly TDD-driven with high-quality
TestRepo-based tests will be rejected immediately.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 249 lines · 4,559 tokens per session scan A ce8e337ce010
git-ai AGENTS.md is an instructions file published in the GitHub repository git-ai-project/git-ai (2,525 stars, last pushed today), licensed Apache-2.0. It adds 4,559 tokens to every session, about $0.0228 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
ontology-atlas AGENTS.md
Instructions for wlsdks/ontology-atlas, covering agents.md — ontology-atlas, product and non-negotiable architecture, start here, structure and routes and operating gates and skills.
ephemeral-sandbox CLAUDE.md
Instructions for Ephemeral-AI-Lab/ephemeral-sandbox, covering claude.md, project, engineering practice (required), build & test and sandbox tools.
matt-skills-with-to-goal CLAUDE.md
Instructions for tt-a1i/matt-skills-with-to-goal: Skills are organized into bucket folders under skills/.
coral CLAUDE.md
Instructions for cdknorow/coral, covering claude.md - coral go, mission, testing, go unit tests and legacy parity tools (historical reference).
pixir CLAUDE.md
Instructions for Ranvier-Technologies/pixir, covering pixir harness - legacy agent guide, current map and commands.
Agent-Memory-Bridge AGENTS.md
Instructions for zzhang82/Agent-Memory-Bridge, covering agent memory bridge contributor instructions, setup and checks, architecture boundaries, mutation and migration invariants and benchmark expectations.