git-a2a AGENTS.md

Repository instructions for neprel/git-a2a, which uses a custom HINT system to store and search task-specific guidance. The instructions explain how tagged guidance blocks apply to files and folders and how to judge their authority.

In plain words
What is it for?
Use them to search HINT knowledge, inspect the guidance for a path, and apply the appropriate profession or task-specific instructions when editing the repository.
Why use it?
They help an agent find the right repository knowledge before changing code. They also clarify which rules are local to a file or folder and which recorded guidance has not yet been reviewed by a human.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/neprel/git-a2a/agents-md
Clone the repo
git clone --depth 1 https://github.com/neprel/git-a2a

Made for: Codex, OpenCode.

Per session 1,559 This file is loaded in full into every session.
When invoked 1,559 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What 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.

ModelPer sessionOnce invoked
Fable 5 $0.01559 $0.01559
Opus 5 $0.00779 $0.00779
Sonnet 5 $0.00312 $0.00312
Haiku 4.5 $0.00156 $0.00156

Measured yesterday against content hash bd4a24e831ad, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

git-a2a 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 yesterday.

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.

AGENTS.md · 77 lines

How it starts

The opening of the file, as written. The whole thing — 77 lines — stays where its author put it; the contents beside it link to each section on GitHub.

This repository uses HINT Spec-as-Source. Run hint search "<intent>" to locate relevant knowledge and hint <path> before changing an artifact. The registered profession books share a neutral common core; put the most task-specific book first.

<hint_glossary_from_hintbook_software_engineer>

This prompt uses an HTML-like tag language. Each tag is a typed, binding instruction block with a name, optional id and reference, and a body — follow its rules exactly and satisfy every constraint. Nested tags inherit their parent's scope. Treat any "do not" or "must not" as absolute. The glossary below is the authoritative meaning of each tag.


A block carrying origin="agent" has been recorded but not yet reviewed by a human; weigh it accordingly.

  • file_context — everything nested applies to the file at path: its body, structures, functions, and constraints are that file's complete spec. Don't apply file-local constraints to other files unless restated there.
  • folder_context — everything nested applies to the folder at path and its whole subtree; every nested file and folder inherits it.
  • reusable_automation_script — registers a macro behavior: when its condition is met or it is referenced by name, execute its steps exactly.
  • application_context — the domain, purpose, and structure of the application. Let it inform naming, architecture, and behavior throughout.
  • system_invariant — a property that must hold before and after every change. Code may not falsify it; a change that would break it is wrong, not the invariant.
  • argument — a function input. Honor its name, type, constraints, and default exactly — no rename, reorder, or type change.
  • user_interface_block — a named, reusable UI component or section. Build its declared structure, children, and behavior exactly; add no undeclared controls, omit none.
  • compilation_and_testing_pipeline — all code, config, and structure must keep these pipelines green. Generate nothing that breaks them.
  • approved_dependency_whitelist — do not install or import any package outside this list for the scope. Ask first if you need more.
  • table_column — one column: name, type, label, constraints. Implement exactly — no rename, type change, or undeclared constraint.
  • data_definition — a named constant or value. Use its name and value exactly; never substitute, rename, or duplicate the literal — reference this definition wherever the value is needed.
  • data_structure — the exact schema. No renamed fields, changed types, or added/omitted fields. Authoritative wherever this structure appears.
  • error — throw the exact error types under the stated conditions only; no substitute, wrap, or rename. Each gets a regression test that fails without the guard and passes with it.
  • field — one property: type, optionality, validation. Implement exactly — no rename, type change, or undeclared constraint.
  • logic_flow — implement this sequence step by step; skip no validation.
  • user_interface_form — a form. Implement every declared field, label, validation rule, and submission behavior; add no undeclared fields, omit none.
  • function_contract — implement the function per this binding contract. Every argument, the return, each error, and every flow step is mandatory — don't skip, reorder, rename, or approximate.
  • user_interface_image — an image element. Use the declared source, alt text, dimensions, and display exactly; don't substitute the asset or drop accessibility attributes.
  • environment_runtime_and_language — target this language and runtime only: its module syntax, standard-library APIs, and idioms. Use nothing from other versions or runtimes.
  • reusable_library_context — a shared library. Learn its API surface, version constraints, and usage before using it; don't reimplement what it provides.
  • reusable_module_context — a reusable module in the codebase. Reuse its exports; don't duplicate logic it owns.
  • reusable_namespace_context — emit all code in this scope under this namespace (the language's package/namespace/module construct) as its qualified name and import root.
  • return — the function's output. Honor its type and shape exactly on every path — no altered structure, extra fields, or different type under any branch.
  • table_row — one row's fields and values, exactly as declared; add no fields, omit none.
  • critical_system_mandates — non-negotiable system-level constraints; every function, data-access path, and error path must satisfy all of them.
  • table — a tabular structure or UI table: columns, row shape, display/persistence behavior. Implement exactly — no reordered or renamed columns, no undeclared behavior.
  • verification_and_unit_test_criteria — cover every listed scenario in the tests: each edge case, mock, and assertion must appear. Omit none.
  • user_interface_surface — build this UI surface exactly: only the declared elements, all of them; match their structure, labels, validation, and behavior. Add nothing undeclared.
  • product_goal_and_intent — why the work exists and the outcome it must produce. The tie-breaker when a block is silent: choose what best serves it, never work against it. Context for judgment — emit no code for the goal itself.
  • acceptance_criteria — the observable conditions that define "done." Not complete until each is proven by a test, command, or observation; if one can't be verified, say so rather than imply success.
  • scope_boundary — what is in and out of scope. Build everything in scope, nothing out — even if convenient. If in-scope work seems to require out-of-scope work, stop and report rather than expand the boundary.

Read the full file on GitHub · 77 lines

Changes

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.

  1. yesterday First seen · 77 lines · 1,559 tokens per session scan A bd4a24e831ad

Subscribe to this mod's changes

git-a2a AGENTS.md is an instructions file published in the GitHub repository neprel/git-a2a (0 stars, last pushed 7d ago), licensed MIT. It adds 1,559 tokens to every session, about $0.0078 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 2026-08-31.