product-delivery-harness
AGENTS.md1 starsChanged 18 days agoMITView on GitHub
Use this suite only when the user explicitly requests a named skill or continues its already requested workflow. A $skill-name invocation or a natural-language request naming the skill qualifies; discussing a skill does not. Required checks and dependencies stay within that requested outcome. A new top-level stage needs an explicit request. Installed skills, seeded instructions, product documents and existing PLAN/RUN files do not activate a workflow. For ordinary tasks, follow effective host and repository instructions plus the user's request. Do…
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 32,671 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 repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- delivery-harnessSkill
- design-system-compilerSkill
- growth-reviewSkill
- product-definition-builderSkill
- ui-design-builderSkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 5
Use this suite only when the user explicitly requests a named skill or continues its already requested workflow.
- Distinctive: line 10
Do not run Harness routing, document sync, Epic/index updates, template audits or PLAN/RUN solely because these instructions exist.
- Distinctive: line 30
Read the complete linked section before its applicable action.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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
Project Rules
Skill Activation
Use this suite only when the user explicitly requests a named skill or continues its already requested workflow.
A $skill-name invocation or a natural-language request naming the skill qualifies; discussing a skill does not.
Required checks and dependencies stay within that requested outcome. A new top-level stage needs an explicit request.
Installed skills, seeded instructions, product documents and existing PLAN/RUN files do not activate a workflow.
For ordinary tasks, follow effective host and repository instructions plus the user's request.
Do not run Harness routing, document sync, Epic/index updates, template audits or PLAN/RUN solely because these instructions exist.
Existing data protection and action authorization still apply.
Runtime Boundary
- This file is the entry. It contains authority boundaries and routes, not complete procedures. Preserve the current host's effective instruction discovery and precedence.
- Within an explicitly requested skill workflow, apply Parent-Owned Delegation before direct or managed routing. Resolve executors and delegation from effective agent configuration, AGENTS instructions and observed tools. The actual frontend author reads the complete pinned
frontend-designSKILL.md and applicable project design skills in its own context; verify the full-tree pin independently. Fresh independent review binds the exact candidate. One writer owns each checkout, including the parent; preserve required independent review and actual execution evidence. - For explicitly requested managed execution, use General Runtime Adapter for observed native capabilities, authorization, isolation and result contracts. Capability does not grant an action.
- Within an explicitly requested
delivery-harnessworkflow, direct work uses one writer, or a Direct Fan-Out of sibling writers, and one coherent verification sequence. Managed rules apply only after work routes into PLAN/RUN. - Preserve unrelated local work. Important local data needs explicit owner approval before deletion, overwrite, or move.
- This source repository intentionally omits the template's Skill Bindings (it has no consumer stage slots), Behavior Specs (it has no feature files), Harness Version (it is the source, not an installed consumer), Deployment section and its Deployment and External Setup routes (it deploys and sets up no product), and the delivery-acceptance bullet (it has no product accounts or data). Its Git Flow replaces the template's Git Safety, and its Required Reading replaces the template's. Treat these as local choices, not drift, in the handoff audit.
Worker Policy
Effective agent definitions and AGENTS instructions own worker selection, models, effort, fallback and delegation topology. Shared skills own methods, inputs, authorization, evidence and acceptance; they do not supply competing worker policy. Persisted RUN labels describe responsibilities, not native agent identities; preserve the pinned execution protocol.
Mandatory Governance Routes
Harness-specific routes below apply only within an explicitly requested skill workflow. Source reading, verification and Git rules still govern maintenance of this repository without starting a consumer delivery workflow. Read the complete linked section before its applicable action. The links are mandatory, and an owner file's rules cannot be widened by this entry.
| Trigger | Complete rule owner |
| Explicit invocation of a skill in this suite | Project Entry |
| Task start; significant edit, commit, merge, branch switch, or observed external change | Repository Change Checkpoints |
| Completion or handoff | Handoff Documentation Audit and Source Completion |
| Local deletion, overwrite, or move | Protect Local Data |
| Implementation or repair | Keep Changes Simple and Source Core Development Principles |
| New or changed documentation | Document Writing |
| A task introduces or retires a local-only file class | Gitignore Contract |
| Product requirements, architecture, UI impact, redesign, pricing, or partner channel | Product Contracts |
| Pricing, paid access, purchase-gated features, or outside sellers | Monetization And Partner Channels |
Editing skills/, a documented flow, or running a seeded consumer flow | Source Required Reading |
| Decomposing PRD work into missions | Mission Task Split |
| Branch, commit, integration, archive, publication, cleanup, or source main landing | Source Git Flow and Atomic Commit Convention |
| Every atomic source task before verification | Required Verification |
| Formal release, version promotion, tag, bundle installation, or host reload | Update Local Skills and Required Verification |
| Review or acceptance claim | Source Review Guidelines |
| Creating or resuming managed PLAN/RUN | Managed And Consumer Delivery, Conditional Project Operating Rules, and the canonical Harness SKILL.md |
The Epic and index record observed work; neither record grants product approval, marks a test PASS without evidence, or authorizes a commit, branch, install, push, deployment, cleanup, publication, or external action.
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.

