meta-architect / templates
AGENTS.md5 starsChanged 7 months agoMITView on GitHub
This repository uses Meta-Architect as a structured orchestration layer for programmatic architecture and verified engineering. The purpose of this contract is to make repository behavior explicit, reviewable, and stable enough to be reused across generated or installed surfaces. The primary trigger surfaces are: Every task is triaged as P0/P1/P2/P3 before mutation. The default execution contract is understand, design, trim, guardrails, and rollout. Durable task state must retain the priority, plan, definition of done, verification evidence, and rollback path. Do not…
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.
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
22 rules. About 0 are common, 13 are distinctive, and 9 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈13
Distinctive · estimate
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
No implementation should proceed as if it were approved unless the required architecture, evidence, logic, and security gates are satisfied.
- Distinctive: line 70
If any required field is missing, red, or unverified, the workflow should stop and report blockers clearly.
- Distinctive: line 76
no other persistent developer branch is supported
- Distinctive: line 93
prefer known upstream repos and official docs first
- Distinctive: line 94
use discovery accelerators to find or narrow candidates
- Very specific: line 100
Prefer repo-specific https://gitmcp.io/{owner}/{repo} endpoints
- Very specific: line 101
Treat https://gitmcp.io/docs as fallback policy only, not a normal approved evidence source for build-unlocking decisions
- Distinctive: line 102
Treat discovery accelerators as discovery-only until a candidate is mapped back to an exact upstream repository and validated through …
- Distinctive: line 116
Use the role that best matches the task:
- Very specific: line 118
Use $arch for system shape, stack options, component boundaries, trade-offs, and blueprinting.
- Very specific: line 119
Use $sage for evidence-backed package, framework, and library selection.
- Very specific: line 120
Use $flow for workflow correctness, state transitions, data flow, and edge-case review.
- Very specific: line 121
Use $vet for security checks, dependency risk review, and release-sensitive blocking findings.
- Very specific: line 122
Use $vibe for developer experience and user experience review.
- Very specific: line 123
Use $build only after required gates are satisfied and only for bounded implementation planning or execution.
- Distinctive: line 127
When working in this repository:
- Distinctive: line 129
prefer explicit reasoning and explicit file changes
- Distinctive: line 130
do not skip validation steps silently
- Distinctive: line 131
do not weaken hooks or gates for convenience
- Very specific: line 132
do not commit runtime .ma state
- Distinctive: line 133
do not treat release-sensitive changes as trivial edits
- Distinctive: line 147
no hidden runtime residue is shipped as source
≈ “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
Meta-Architect Operating Contract
This repository uses Meta-Architect as a structured orchestration layer for programmatic architecture and verified engineering. The purpose of this contract is to make repository behavior explicit, reviewable, and stable enough to be reused across generated or installed surfaces.
Core triggers
The primary trigger surfaces are:
$arch— architecture and stack blueprinting$sage— evidence-backed OSS discovery through MCP/GitMCP collections$flow— business logic and state-transition review$vet— security and risk review$vibe— DX/UX review$build— bounded implementation planning and build execution handoff
Senior execution default
Every task is triaged as P0/P1/P2/P3 before mutation. The default execution contract is understand, design, trim, guardrails, and rollout. Durable task state must retain the priority, plan, definition of done, verification evidence, and rollback path. Do not treat a successful local command as completion without the required checks and receipts.
Core rule
No implementation should proceed as if it were approved unless the required architecture, evidence, logic, and security gates are satisfied.
External review findings rule
Every finding reported by the chatgpt-codex-connector must be reviewed, implemented or explicitly resolved with evidence, and re-checked before the change is considered complete. Do not dismiss, skip, or claim resolution for a finding until the affected behavior is verified by the smallest appropriate test, check, or reproduction. A pull request remains unresolved while any connector finding lacks a recorded resolution and fresh verification evidence.
Kernel vs extension model
Meta-Architect should be treated as a core system plus extension surfaces:
- Core system
- state and decision files
- gate enforcement
- MCP endpoint policy
- CLI orchestration commands
- Extension surfaces
- publishable skill folders
- plugin bundle
- missions and demos
- prompts and templates
The core must stay stable and inspectable. Extensions may vary, but they must not weaken the kernel contract.
Gate rules
The repository recognizes the following status fields:
idea_statusarchitecture_statusevidence_statuslogic_statussecurity_statusexperience_statusbuild_statusmerge_statusrelease_status
Build lock rule
$build must remain locked unless all of the following are true:
idea_status = CLEARarchitecture_status = APPROVEDevidence_status = VERIFIEDlogic_status = GREENsecurity_status = GREENexperience_status = GREENorWAIVED
If any required field is missing, red, or unverified, the workflow should stop and report blockers clearly.
Merge and release rule
- human work is completed on
dev - release promotion is allowed only from
devtomain - no other persistent developer branch is supported
Evidence rule
Major technology recommendations must be supported by evidence from configured MCP/GitMCP sources. Do not invent fake repositories, fake endpoints, or unsupported claims.
Approved discovery accelerators include:
- Ossium (
https://ossium.live/home) for trending OSS, curated repos, YC-backed repos, GSoC orgs, and contribution leads - Trendshift (
https://trendshift.io/) for rising GitHub engagement and topic momentum - Dev Hunt (
https://devhunt.org/) for newly launched developer tools - Libraries.io (
https://libraries.io/) for package/dependency metadata, with caution because its public data is scraped and not validated/curated for accuracy - Open Hub (
https://openhub.net/) for project activity, contributor, popularity, and comparison signals - Open-source Projects (
https://www.opensourceprojects.dev/) for curated OSS discovery and detailed project writeups
These discovery accelerators do not replace upstream repo evidence or official docs for approval decisions.
Canonical evidence order:
- prefer known upstream repos and official docs first
- use discovery accelerators to find or narrow candidates
- map selected candidates back to exact upstream repositories
- approve only after upstream repo and official-doc verification
Exact endpoint rule
- Prefer repo-specific
https://gitmcp.io/{owner}/{repo}endpoints - Treat
https://gitmcp.io/docsas fallback policy only, not a normal approved evidence source for build-unlocking decisions - Treat discovery accelerators as discovery-only until a candidate is mapped back to an exact upstream repository and validated through approved evidence paths
Logging rule
Architecture decisions, evidence, audits, and release-relevant outcomes should be written to repository-visible local files where applicable, especially:
.ma/decisions.json.ma/release.json.ma/evidence/sources.json.ma/evidence/audits.json.ma/evidence/cves.json
Skill routing
Use the role that best matches the task:
- Use
$archfor system shape, stack options, component boundaries, trade-offs, and blueprinting. - Use
$sagefor evidence-backed package, framework, and library selection. - Use
$flowfor workflow correctness, state transitions, data flow, and edge-case review. - Use
$vetfor security checks, dependency risk review, and release-sensitive blocking findings. - Use
$vibefor developer experience and user experience review. - Use
$buildonly after required gates are satisfied and only for bounded implementation planning or execution.
Prompt behavior
When working in this repository:
- prefer explicit reasoning and explicit file changes
- do not skip validation steps silently
- do not weaken hooks or gates for convenience
- do not commit runtime
.mastate - do not treat release-sensitive changes as trivial edits
Contributor expectation
Changes to skills, prompts, package metadata, workflows, release docs, or plugin surfaces are release-sensitive. These changes should be tested, documented, and described clearly in pull requests.
Release expectation
Production release means:
- the core trigger flow works end-to-end
- release artifacts are validated
- skills packaging and installation succeed
- docs reflect actual behavior
- no hidden runtime residue is shipped as source
Related surfaces
CONTRIBUTING.mddocs/release-spec.mdtemplates/catalog-manifest.jsonprompts/skills/
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.

