forge
AGENTS.md56 starsChanged 8 months agoMITView on GitHub
The separation between agents is deliberate. The reviewer has fresh context and no implementation bias. The verifier never sees execution details, only checks outcomes against the spec. Three skills are inlined into agent definitions and run automatically. No manual invocation required. All three degrade gracefully. If no graph or DESIGN.md exists, agents behave exactly as before v0.2.0.
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 repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- backpropagationSkill
- brainstormingSkill
- caveman-internalSkill
- collaboratingSkill
- design-systemSkill
- executingSkill
- graphify-integrationSkill
- karpathy-guardrailsSkill
- planningSkill
- reviewingSkill
The file
We split this agents.md into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 25
All three degrade gracefully.
≈ “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.
Files
Seven Specialized Agents
| Agent | Role | Min Model | Key Constraint |
| forge-speccer | Writes R-numbered specs with testable criteria | sonnet | One question at a time, capability-aware criteria |
| forge-planner | Decomposes specs into streaming DAGs | sonnet | Coverage verification, no gold-plating |
| forge-executor | Implements tasks with TDD + convention inference | haiku | Worktree + checkpoints, follow existing patterns |
| forge-researcher | Multi-source research before implementation | haiku | Produces reports only, never writes code |
| forge-reviewer | Two-pass review: spec compliance + blast radius | sonnet | Caveman for minor, verbose for security |
| forge-verifier | Four-level goal-backward verification | sonnet | Caveman for pass, verbose for gaps |
| forge-complexity | Scores task difficulty across 5 dimensions | haiku | Lightweight, runs on every command startup |
The separation between agents is deliberate. The reviewer has fresh context and no implementation bias. The verifier never sees execution details, only checks outcomes against the spec.
Cross-Cutting Skills (v0.2.0)
Three skills are inlined into agent definitions and run automatically. No manual invocation required.
| Skill | Used By | Behavior |
| karpathy-guardrails | executor, reviewer, planner | Four behavioral principles: Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution. Executor self-checks; reviewer flags violations as IMPORTANT. |
| graphify-integration | planner, researcher, reviewer, executor | Auto-detected via graphify-out/graph.json. Planner aligns tasks with community clusters. Researcher queries graph before external docs. Executor gets focused context. |
| design-system | brainstorming, planner, executor, reviewer | Auto-detected via DESIGN.md. Planner tags UI tasks. Executor loads design tokens. Reviewer runs design compliance pass. |
All three degrade gracefully. If no graph or DESIGN.md exists, agents behave exactly as before v0.2.0.
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.

