agentleFS
GerardoRdz96GerardoRdz96/claude-skills/skills/framework-builder/SKILL.md

framework-builder

Skill0 starsChanged 4 months agoMITView on GitHub

Turns Gera's thinking or a topic into a reusable DECISION FRAMEWORK — a named decision rule / rubric / scorecard / gate that terminates in an ACTION (pick / build / skip / score), in the 3Ms / Four Cs / routine-placement shape. Type `/framework-builder` to structure a decision into an executable tool, often downstream of /grill-me. Not for software/code frameworks, prose SOPs, or wiki pages with no decision rule.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill 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
  1. Distinctive:

    1. Gather: Read the capture / named pages, or ask the short targeted question set.

    line 28

≈ “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

What this skill does

Turns Gera's thinking or a topic into a reusable, structured framework — and graduates it into the wiki, wiring it into Servy when it's a decision rule Servy runs. It is the synthesis step that produces the kind of page Servy already keeps (3Ms, Four Cs, the routine-placement framework, ai-applicability): a named, ordered, executable decision tool.

The one non-negotiable: a framework's output is an ACTION (pick / build / skip / score / which-home), not understanding. If what you'd produce is descriptive (the reader finishes with thoughts but no verdict), it is a wiki page, not a framework — stop and route it to normal references/ ingest. This test gates the whole skill.

Full template, the 11-point quality gate, antipatterns, the two type patterns, graduation, and the wiring rule live in reference.md (read it before building).

Where it sits (don't collide)

  • /grill-me EXTRACTS raw thinking to a capture. /framework-builder is downstream: it STRUCTURES a capture (or a short focused extraction it runs itself) into a decision tool. If the thinking isn't captured yet and is deep, offer to run /grill-me first, then chain here.
  • brainstorming explores NEW design space; /skill-builder builds skills. This builds a framework (a reusable decision rule), which may then back a skill.
  • NOT for: software/code frameworks, prose playbooks or SOPs, or ordinary wiki pages that have no decision rule (those are normal references/ ingest).

Process

0. Scope (decide before gathering).

  • Type: personal thinking/decision framework (client = Servy/Gera) or generic client/Jumpstart delivery framework (client = a customer's team)? They share the skeleton but differ in grounding + wiring — see reference.md "Two types".
  • Source: a fresh topic (run a SHORT focused extraction — a handful of targeted questions, not a full grill) or existing material (a named knowledge/grills/ capture, named wiki pages, a decision log) to synthesize. Optional: research-ground it (/deep-research or web) when the claims need external authority (as routine-placement did).
  • SoftServe boundary (any client/Jumpstart work) + data-classification gate. Generic methodology ONLY — anonymized archetypes. HARD STOP (does NOT enter Servy): customer IP/data/names, SoftServe Internal+ methodology, engagement artifacts, proprietary runbooks, or "sanitized" customer-derived material unless explicitly approved. And before any step routes to an external tool (/deep-research, web, Codex, the multi-brain in step 1/4): classify the inputs — never send SoftServe/customer/Internal+ material to external tools; those are personal/public-data only (references/softserve-ai-usage-policy.md).

1. Gather. Read the capture / named pages, or ask the short targeted question set. Pull the raw decisions, ratios, stop-conditions, and real examples. Don't re-implement grill-me; lean on it.

2. Find the axis + draft the decision rule. Name the ONE optimization axis / load-bearing fact (the bolded scoping line). Draft the decision rule in the fitting form — ordered-rules / flowchart / gated-table (reference.md). Make every step emit a named branch/verdict; state order as a rule.

3. Complete the instrument. Add the verdict/scoring rubric (gated if ANY / only when ALL + tie-breakers + "guide not arithmetic"), the stop-rules/tells (so it can say no to itself), and ≥2 dogfooded applications where possible — one ACTION case AND one STOP/boundary case (a framework only ever used to say "yes" is unvalidated). If only one real case exists, file it as provisional, or keep it as normal wiki ingest until a second case validates it.

4. Quality gate (MANDATORY, before graduating). Append an explicit pass/fail checklist to the draft — each of the 11 gate items in reference.md marked ✓/✗, naming the terminal verdicts, the stop-rules, and the dogfood result(s). If ANY item is ✗ — especially the non-negotiable (all prose / no executable decision rule) — STOP: fix it or route to normal wiki ingest. Do not file a fake framework. Codex review is MANDATORY (not just "substantial") for any framework that will be wired into Servy or used for client/Jumpstart delivery (No-Self-Review Law): git diff | codex exec --skip-git-repo-check "Is this a real, executable decision framework or descriptive prose? Find any step that doesn't branch, any verdict decided by vibes, any missing stop-rule, any ungated 'it depends'."

5. Graduate (wiki-protocol). Write references/<kebab>-framework.md to the skeleton; update references/index.md (summary == the page's one-liner); register a knowledge/ raw source in knowledge/README.md if distilled from one (mark "(digest)"); append ## [<date>] create | <Name> to references/log.md; lay **Related:** cross-links (no orphans); run the build-triage reflex; optionally fire warden. Full checklist in reference.md.

6. Wire into Servy (decision frameworks only — PROPOSE-THEN-CONFIRM). If it's a rule Servy runs repeatedly, propose the exact wiring (skill gate / one-line CLAUDE.md pointer / standing hook), show Gera the diff, and wait for an explicit yes — never auto-edit CLAUDE.md or a SKILL.md, never touch identity-core, respect the 200-line cap. Wiring must be bidirectional (page names the skill AND the skill/CLAUDE.md names the page). The page's own ## Wiring section can ship without confirmation (it's wiki prose). Client/Jumpstart frameworks get NO CLAUDE.md standing behavior and NO automatic behavioral gate. Cross-linking a delivery skill (/jumpstart, /forge) so it consumes the page is itself wiring — so it's also propose-then-confirm + bidirectional, not a free edit.

7. Optional HTML artifact. Offer a polished HTML artifact (artifacts/frameworks/<date>-<slug>.html) per references/html-as-llm-output.md — a framework with a flowchart/table is a strong candidate to forward.

Output

  • Canonical: references/<kebab>-framework.md (markdown, no frontmatter) + references/index.md entry + references/log.md line.
  • Optional: an HTML artifact; a proposed CLAUDE.md/skill wiring diff (applied only on Gera's yes).

Guardrails

  • The decision-rule gate is non-negotiable — no executable rule, no framework (route to normal ingest).
  • SoftServe boundary — client frameworks are generic-only; customer IP never enters Servy.
  • Don't duplicate /grill-me — chain off it for deep extraction; this skill does only a short focused extraction itself.
  • Wiring is propose-then-confirm + bidirectional — never auto-edit CLAUDE.md/identity-core; respect the budget cap.
  • No orphans, honest provenance — index entry, ≥1 Related link, a Sources line, dated footer (see reference.md).
  • This skill is conversational/synthesis — stays in the live session; it may fan out to /deep-research or Codex, but doesn't fork itself to an isolated agent.

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.