Skill

Practical execution workflows and creative thinking lenses for engineering, research, and design.

38 entrypoints across 6 suites: 20 SOPs, 15 heuristics, and 3 auxiliary routers.

Two kinds of skill

• SOP — carry out a chosen task. Inputs, steps, branches, concrete artifacts, verification, and completion conditions. A run-based CLI records progress where execution constraints are useful.
• Heuristic — open possibilities. Philosophical perspectives, analogies, associations, inversions, and thought experiments. These are not multi-angle diagnostic checklists. They have no mandatory gates, certificates, scores, or obligation to end with an action plan.

The human chooses what is worth pursuing. Exploration can hand a selected direction to an SOP, but it can also end with a sketch, a new question, or several unresolved possibilities. Routers help choose a skill—or a direct answer. Reference essays and design probes are supporting resources, not more skill categories.

Suites

Start with engineering:groundwork for an unclear task, engineering:assay for a regression, surface:bearings for alternative frontend directions, or atelier:canon for visual possibilities. The three pilot routers are available in engineering, surface, and atelier. There is no mandatory suite-wide pipeline.

Setup

In Claude Code, add the repository marketplace and install only the suites you want:

/plugin marketplace add IamK77/Skill
/plugin install engineering@skill
/plugin install distributed@skill
/plugin install inquiry@skill
/plugin install quarry@skill
/plugin install surface@skill
/plugin install atelier@skill

Invoke a plugin skill by its namespace, such as /engineering:assay or /atelier:color. Alternatively, copy skills/// into your host’s skills directory, preserving references, LICENSE, and NOTICE. A pilot router also needs the sibling skills it links to: install its suite, or preserve that suite’s sibling layout when copying it manually. Copying pilot/ alone does not install its destinations. Heuristics and routers need no checklist runtime.

The rewritten SOPs require checklist 0.5 or newer. The source and committed bundle in this working tree implement that interface; this does not imply an npm release has been published. With Node 20 or newer, the repository’s self-contained bundle needs no dependency install:

node /absolute/path/to/Skill/devtools/checklist/bundle/checklist.mjs --help

Use that invocation wherever a skill says checklist, or put the matching local CLI on PATH. Installing a suite plugin does not by itself install a global CLI. Do not use an older global command without checking its version. See the CLI guide and migration notes.

What checklist records—and what it does not prove

Each new task gets a distinct run ID, even when skill and project are the same. The project path is fixed at creation. Loading a skill never clears state; resuming is explicit.

checklist init /path/to/skill --new --path /path/to/project
# Retain the returned ID as RUN; follow that skill's concrete item names.
checklist show --run "$RUN"
checklist resume "$RUN"
checklist report --run "$RUN"

• check records a manual confirmation and its stated evidence. It cannot establish whether the claim is true or whether a person consented.
• verify executes the defined sensors and records their readings. Success covers those commands, not every manual item or the whole stage.
• advance explicitly closes a fulfilled phase. Later phase mutations require earlier phases to be closed.
• done completes a fulfilled run and retains its history. reset --reason ... abandons a run; it is not an alias for completion and does not erase evidence.
• N/A is allowed only where declared and requires a reason. Failed, errored, pending, and stale results do not satisfy completion.

The CLI constrains its own state transitions. It cannot stop an agent from reading later prose, doing work outside the CLI, or making a false manual assertion. It is not a sandbox, an authorisation system, or a proof of engineering quality.

Maintained scope

Active work here is the skills, checklist CLI, authoring checks, and design probes. The root A/B experiment platform has been removed from the tracked tree; its source and artifacts remain in Git history. The supported quarry:touchstone repository-assessment skill remains available.

The reference library follows the same split as the entrypoints: SOP method cards and creative lenses with optional technique notes. The full migration rewrote 194 references and retained 4 already-current routing notes; it does not leave a hidden legacy workflow behind. See AUTHORING.md and the migration record.

Development checks

cd devtools/checklist
npm ci
npm run bundle:check   # fails on a stale committed artifact; never rewrites it
npm test              # builds source; same run contract tests compiled and relocated bundle
cd ../..
python3 devtools/skill-lint.py
python3 -m unittest discover -s devtools -p 'test_*.py'
node devtools/checklist/dist/index.js lint skills --strict

After intentional CLI changes, run npm run bundle and review the generated diff before the freshness check. Probe tests remain in canon and color.

License

Apache-2.0, Copyright 2026 IamK77. See LICENSE and NOTICE.