agentleFS
DojoGenesisDojoGenesis/openclaw-plugin/skills/poc-tool-forge/SKILL.md

poc-tool-forge

Skill0 starsChanged 9 months agoMITView on GitHub

End-to-end pipeline to turn an idea into a live single-page TresPies POC tool — source/pick an idea (Build Forge brief or operator concept), build it to the house design system, browser-verify, deploy to *.trespies.dev, and document. Use when the operator says "build a POC tool", "make something from the brief", "ship a new trespies-poc tool", "spin up a landing demo", "check trespies-poc for ideas", or "/poc-tool-forge". Orchestrates the house-poc-builder, static-tool-verify, and cf-pages-trespies-deploy skills.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 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.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    All paths absolute. trespies-poc/ is its own git sub-repo (local-only unless a remote is added) — commits there don't touch the parent.

    line 19
  2. Very specific:

    Default idea source: the nightly Build Forge drop — trespies-poc/build-forge/briefs/LATEST.md.

    line 23
  3. Very specific:

    Leave uncommitted unless asked.

    line 61
  4. Distinctive:

    Real data > fabricated: when a credible source exists (the ip-scout / contagion pattern): capture via an MCP, bake a snapshot with full …

    line 68

≈ “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.

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

OpenClaw Integration: This skill is invoked by the Dojo Genesis plugin via /dojo run poc-tool-forge. The agent receives project context automatically via the before_agent_start hook. Use dojo_get_context for full state, dojo_save_artifact to persist outputs, and dojo_update_state to record phase transitions and decisions.

poc-tool-forge — idea → live trespies.dev tool

The repeatable pipeline that produced quorum / counterclock / body-almanac / phenotype-draft / contagion / bodyglossary (all live on *.trespies.dev). Each POC is a self-contained single-page index.html under trespies-poc/<slug>/, deployed as its own Cloudflare Pages project. This skill is the conductor; it delegates to three phase skills: house-poc-builder · static-tool-verify · cf-pages-trespies-deploy.

All paths absolute. trespies-poc/ is its own git sub-repo (local-only unless a remote is added) — commits there don't touch the parent. Commit/push only when the operator asks.

Phase 1 — Source & pick the idea (don't over-survey; recommend, then confirm)

  • Default idea source: the nightly Build Forge drop — trespies-poc/build-forge/briefs/LATEST.md. Its "Tonight's 5 standouts" are the curated best; the 48-idea list (8 families) is the deep menu. Scores are [u c s ce f] = uniqueness · cashflow · scalability · ceiling · founder-amplification.
  • Present 2–4 candidates with a one-line hook + scores and a clear recommendation, then confirm with AskUserQuestion (the which-idea choice is the operator's and changes everything). Filter for single-page-demoable: no real backend, demonstrable client-side, sharp screenshot, variety against what's already shipped (see trespies-poc/README.md table).
  • Carry forward sensible defaults rather than re-asking every time: demo-ready single page → deploy to *.trespies.dev has been the standing choice this whole session. State the assumption; let them override.
  • The operator may pick several at once ("build the first three") — that's the norm; fan out.

Phase 2 — Build (parallel, one agent per tool)

  • Invoke house-poc-builder per tool. Dispatch one Agent per tool, in parallel (separate folders = no file conflict). Never isolation: "worktree" for file-writing agents (writes get discarded). Model split: Opus for the flagship / most complex; Sonnet for the rest (~1 Opus : 2 Sonnet).
  • Give each agent the full house-design spec, the specific mechanic, the load-bearing guardrail, and its exact output path. Agents start cold — hand them everything (read paths, not memory).

Phase 3 — Verify (independent; never trust agent self-reports)

  • Invoke static-tool-verify per tool: structural + node --check, then a real browser pass (preview_start → exercise the core mechanic → assert zero console errors → screenshot).
  • Fix the bugs a syntax check misses. Real examples caught this session: a "Summer" almanac citing a "March" driver (season-incoherent vocab) and a public-health page labeling a quiet summer as "HIGH" (severity banded on cross-state rank instead of absolute %-of-peak). These pass node --check but fail a human — catch them before deploy.

Phase 4 — Deploy

  • Invoke cf-pages-trespies-deploy per tool: create project → deploy → attach slug.trespies.dev → create the proxied CNAME → verify. Kick off a background watcher until all go 200; don't block.

Phase 5 — Document & track

  • Per tool: a trespies-poc/<slug>/README.md (house pattern — tagline, what-it-is, guardrail, Live URLs, deploy/run). Add a row to the trespies-poc/README.md index table. Register a .claude/launch.json entry (python3 -m http.server <port> --directory trespies-poc/<slug>).
  • Optional ecosystem tie-ins: mirror the build as a Linear issue (the brief tracks builds as OCH-*); if cross-machine parity matters, write a handoff (see cf-pages-trespies-deploy for the token-parity handoff precedent).
  • Leave uncommitted unless asked. When asked: conventional commit in the trespies-poc sub-repo.

Intelligence notes (what makes this not just a checklist)

  • Altitude: recommend, don't dump. One AskUserQuestion at the real fork, then act.
  • Guardrail is load-bearing, not decoration — these deploy publicly and many are health-adjacent. Health/medical-adjacent tools are general-wellness / decision-literacy, never medical advice; the disclaimer banner + footer ship in every build (enforced in house-poc-builder).
  • Real data > fabricated when a credible source exists (the ip-scout / contagion pattern): capture via an MCP, bake a snapshot with full provenance, never invent a number, and sanity-check the rendered framing isn't misleading.
  • Don't poll in the foreground — background-watch domains; the harness re-invokes on completion.

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.