agentleFS
sm-wankosm-wanko/Human-on-exception/AGENTS.md

Human-on-exception

AGENTS.md1 starsChanged 46 days agoMITView on GitHub

Target agents: Cursor / Claude Code / CodeX. Human-facing review and response language follows the language policy. Feature index: docs/testing/core-features.md → docs/ai/packs/ (paths + Flow IDs + completion make commands only; do not copy flow bodies into packs). Apply the AI execution contract to every stage. Humans own Intent / Scope / Answer / Risk acceptance. AI owns execution. Human code review is not a required gate. Follow the language policy. Agent-facing rules are English. Human-facing decision material uses the…

  • 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 33,315 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 repository3 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Follow the language policy.

    line 13
  2. Distinctive:

    If the task changes purpose, terminology, aggregation units, actor boundaries, or evidence meaning, read the relevant concept, active ADR, …

    line 23
  3. Very specific:

    Rules: always read docs/rules/backend-quick.md / frontend-quick.md for touched BE/FE areas.

    line 27
  4. Very specific:

    Follow reading order 0→5. Read one relevant flow and one relevant pack, not the whole repository by default.

    line 41
  5. Very specific:

    Every created Issue / PR must include Pack, Flow ID, allowed paths, and completion make commands.

    line 42
  6. Distinctive:

    Before completion, run the verification level in ai-workflow.md §4 for that PR's base.

    line 43
  7. Very specific:

    Use make or the app's existing commands for verification.

    line 44
  8. Distinctive:

    Do not delete existing implementation or tests based only on docs.

    line 63

≈ “Common” is an estimate: at least 1 in 5 of the 42,283 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AI Agent Rules (Human-on-Exception)

Target agents: Cursor / Claude Code / CodeX. Human-facing review and response language follows the language policy.

Feature index: docs/testing/core-features.md → docs/ai/packs/ (paths + Flow IDs + completion make commands only; do not copy flow bodies into packs).


Execution and decisions

Apply the AI execution contract to every stage. Humans own Intent / Scope / Answer / Risk acceptance. AI owns execution. Human code review is not a required gate.

Follow the language policy. Agent-facing rules are English. Human-facing decision material uses the selected human language by default; do not duplicate translations unless explicitly requested or required by the project.

Review guidelines

When Codex is invoked as a PR reviewer, apply the independent AI review contract in full. The reviewer is not the implementation agent described below; use a separate context and independently inspect accepted Answers / Issue AC, current-state SoT, applicable concept / active ADR / Policy, diff, and tests. Required: identify affected SoT/docs and test artifacts for every behavior/contract change and detect missing updates including files absent from the PR diff. Do not mistake the current-state SoT priority for permission to override the agreed target contract. Do not accept the implementer's completion claim as evidence.

Minimal reading order

  1. Core feature bundle: docs/testing/core-features.md — index of the relevant pack and Flow IDs.
  2. Issue / PR — Flow ID, allowed paths, prohibitions. Before an Issue exists, identify the relevant bundle from the request.
    • If the task changes purpose, terminology, aggregation units, actor boundaries, or evidence meaning, read the relevant concept, active ADR, and Policy first according to domain decision inheritance.
  3. Exactly one relevant flow: docs/flow/<feature>.md — Scope and matrix rows. Read only the relevant 4-point docs when needed.
  4. Corresponding tests: *-SYS-* → apps/backend/tests/System/; *-FE-* → apps/frontend/src/**/*.contract.test.ts / *.integration.test.tsx.
  5. Implementation files: manifest in the Issue / PR or docs/ai/packs/<bundle>.md.
  6. Rules: always read docs/rules/backend-quick.md / frontend-quick.md for touched BE/FE areas. Read detailed conventions only when needed.

Avoid during initial exploration

  • .trash/** — local retired example artifacts only; never use as active Source of Truth or normal discovery input
  • coverage/docs/* generated output
  • cross-repository grep of every file under docs/flow/
  • full TECH_STACK.md / README.md unless the relevant section is needed
  • reading every file in docs/rules/*; open only rules relevant to the touched area

Required for every task

  1. Follow reading order 0→5. Read one relevant flow and one relevant pack, not the whole repository by default. In greenfield state, zero bundles/flows/tests is valid; do not search .trash/ to manufacture current facts.
  2. Every created Issue / PR must include Pack, Flow ID, allowed paths, and completion make commands.
  3. Before completion, run the verification level in ai-workflow.md §4 for that PR's base. Add PHPDoc / JSDoc to touched exported/public surfaces.
  4. Use make or the app's existing commands for verification. Do not make an ad-hoc environment the source of truth. Evidence applies only to the head that produced it.

Completion requirements

Inspect the PR base. Do not hard-code develop or main.

HeadWhenRecord for that head
ChildPR base is an Epic work branchtargeted tests that prove the Issue AC, including that Flow's system tests when the change persists data or crosses a boundary; make lint; make survey when the change touches a route, a flow test matrix, or flow / ui / validation / db structure; independent review of that child PR
IntegrationPR base is the branch that receives completed work: one Epic, several Epics in one PR, or a single Issue with no Epicfull make test, full make survey, and a separate-context review of that head

Child results do not prove the integration head. Several Epics in one PR repeat the integration row on the combined head. A later Epic already verified on a head that contains the earlier Epic does not repeat that same head.


Source of Truth when current-state sources conflict

  1. implementation → 2. system test → 3. docs/flow → 4. generated coverage/*

Do not delete existing implementation or tests based only on docs. This order is for determining current-state facts. Once a change is agreed, the target contract comes from Answers / Issue AC according to AI execution contract §6.

Current facts live in code + the 4-point set (flow / ui / validation / db). Decision history lives in docs/testing/questions/ and docs/testing/adr/. Do not put decision history into the 4-point set.

Concept holds purpose and core concepts. ADR holds active design decisions. Policy holds agreed repeatable criteria. AI maintains and verifies the human-facing 4-point feature overview; human code review and document approval are not normal gates.


Prohibited without explicit request / decision

  • changing API parameters or response contracts
  • adding, removing, or changing business logic
  • removing authentication, CORS, or security configuration
  • rewriting the layer structure or adding business logic to Laravel bootstrap / route wiring
  • treating an external boundary marked N/A in a flow as completed

Rules to open when needed

Touched areaRule
Laravel Backendbackend-quick.md always; backend-coding-conventions.md only for details
Frontendfrontend-quick.md always; frontend-coding-conventions.md only for details
flow / docs synchronizationdocs-follow.md
N/A declarationsdocs-na-conventions.md
SIT / FE Contractsystem-test-strategy.md, frontend-flow-contract.md
member-facing mutation auditaudit-ui-persistence.md
human-facing language / translationlanguage-policy.md

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.