forge-craft
Skill2 starsChanged 6 days agoMITView on GitHub
Use before calling work done: what would its user notice?
In this repository99 other files it gives its agents.
- docs/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- brainstormingSkill
- claude-md-improverSkill
- dispatching-parallel-agentsSkill
- executing-plansSkill
- finishing-a-development-branchSkill
- forge-a11ySkill
- forge-agent-reportSkill
- forge-agentSkill
- forge-apiSkill
- forge-authSkill
- forge-autoresearchSkill
- forge-botsSkill
- forge-brainstormSkill
- forge-cliSkill
- forge-cmsSkill
- forge-code-reviewSkill
- forge-contextSkill
- forge-councilSkill
- forge-databaseSkill
- forge-dataSkill
- forge-debugSkill
- forge-deeplearnSkill
- forge-devopsSkill
- forge-docdriftSkill
- forge-doctorSkill
- forge-ecc-adapterSkill
- forge-ecommerceSkill
- forge-electronSkill
- forge-extensionSkill
- forge-figmaSkill
- forge-fullstackSkill
- forge-gameSkill
- forge-graded-verifySkill
- forge-guardianSkill
- forge-heartbeatSkill
- forge-i18n-docsSkill
- forge-intakeSkill
- forge-integrationSkill
- forge-long-form-writingSkill
- forge-mcp-clientsSkill
- forge-memorySkill
- forge-migrationSkill
- forge-mindmapSkill
- forge-mlopsSkill
- forge-mobileSkill
- forge-n8nSkill
- forge-nightshiftSkill
- forge-observabilitySkill
- forge-paymentsSkill
- forge-plan-largeSkill
- forge-prdSkill
- forge-predictionSkill
- forge-projectbrainSkill
- forge-prompt-coachSkill
- forge-qualitySkill
- forge-ragSkill
- forge-registrySkill
- forge-releaseSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Before any deliverable is called done, handed over or shown to the owner: pages, copy, code, workflows, documents, spreadsheets, reports, …
- Very specific: line 31
Add --json for machine output.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
forge-craft — the craft lens
Owner rule (Intelligence bar): before "done", ask who will look at this or work with it, and what they would notice. Sloppy output is a defect, not a style: fix it, or say plainly what is still weak.
When to run
- Before any deliverable is called done, handed over or shown to the owner: pages, copy, code, workflows, documents, spreadsheets, reports, games, CLIs, APIs, voice or chat agents.
- By the Boss that judges it: Verify Boss (next to the domain gate), Review Boss (stakeholder pass before sign-off), UI Boss (self-check before handing over). A builder may run it on its own work first; that never replaces the independent pass.
Commands
node .claude/forge-bin/forge-craft.cjs lens --list # the lenses and their check counts
node .claude/forge-bin/forge-craft.cjs lens --type web-ui # stakeholders, first ten seconds, checks, sloppy defects
node .claude/forge-bin/forge-craft.cjs plan --deliverables "web-ui,copy-and-content" # combined plan for a mission
node .claude/forge-bin/forge-craft.cjs plan --domain website --text "<request>" # domain lenses + what it names
node .claude/forge-bin/forge-craft.cjs review-template --type web-ui # the self-critique form to fill
node .claude/forge-bin/forge-craft.cjs record --run <id> --type web-ui --file review.json # filled form -> run evidence
node .claude/forge-bin/forge-craft.cjs check --run <id> # recorded reviews still valid?
node .claude/forge-bin/forge-craft.cjs validate # schema + citations of every lensAdd --json for machine output. An unknown type gets its closest lenses plus generic, and says so.
Without Bash (Review Boss), Read the lens file itself: .claude/config/quality/lenses/<type>.json.
How to run a craft pass
- Pick the deliverable types; use
plan --domainwhen unsure. Mixed work gets several lenses. - First ten seconds: open or use the deliverable cold, as each stakeholder, and write what you notice before you read a single check.
- Work the block checks first, then the warn checks. Collect the evidence each check names: a screenshot that was read (not only captured), a command with its exit code, a rendered message, a recomputed figure.
- Fill
review-template: per stakeholder what they notice in ten seconds and on closer inspection; per check a mark, the evidence path, the finding, and the fix or the stated weakness. - Fix what you can now and inspect again. What stays weak goes into the report in plain words: what it is, who would notice, why it is not fixed now.
Marks and rules
- Marks: CLEAN (inspected, evidence attached, nothing a stakeholder would notice) · DEFECT · NOT_INSPECTED · N/A (with the reason).
- A block check marked DEFECT or NOT_INSPECTED means the deliverable is not done. A warn DEFECT is fixed or listed as a stated weakness.
- The tool never marks anything; only a reviewer with evidence does. Never mark CLEAN from reading code when the check asks for a screenshot or a run.
- A STALE LENS notice (a citation no longer matches its source) is reported, not ignored.
Lenses
web-ui, copy-and-content, code, automation-workflow, documents-and-spreadsheets, data-and-reports, game-and-interactive, cli-and-devtools, api-and-integration, voice-and-chat, generic. Each check says whether it is cited (a playbook, a hard rule, an agent checklist or the owner's own evidence) or professional practice (with a one-line reason).
References
references/review-form.md: how to fill the form well, with a worked example.references/lens-authoring.md: how to add or change a lens without padding or invented rules.
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.

