Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Semantic workflow composer — parse natural language workflow description into a DAG of skill/CLI/agent nodes, auto-inject checkpoint save nodes, confirm with user, persist as reusable JSON template. Triggers on "wf-composer " or "/wf-composer".
.claude/skills/wf-composer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | — | — |
| case-09 | ✗→✓ | ▲ Improved | — | — |
| case-03 | ✗→✓ | ▲ Improved | — | — |
| case-21 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
Parse user's semantic workflow description → decompose into nodes → map to executors → auto-inject checkpoints → confirm pipeline → save as reusable workflow-template.json.
User describes workflow in natural language
-> Phase 1: Parse — extract intent steps + variables
-> Phase 2: Resolve — map each step to executor (skill/cli/agent/command)
-> Phase 3: Enrich — inject checkpoint nodes, set DAG edges
-> Phase 4: Confirm — visualize pipeline, user approval/edit
-> Phase 5: Persist — save .workflow/templates/<name>.json| Constant | Value | |----------|-------| | Session prefix | WFD | | Template dir | .workflow/templates/ | | Template ID format | wft-<slug>-<date> | | Node ID format | N-<seq> (e.g. N-001), CP-<seq> for checkpoints | | Max nodes | 20 |
Parse $ARGUMENTS.
| Detection | Condition | Handler | |-----------|-----------|---------| | Resume design | --resume flag or existing WFD session | -> Phase 0: Resume | | Edit template | --edit <template-id> flag | -> Phase 0: Load + Edit | | New design | Default | -> Phase 1: Parse |
Resume design session:
.workflow/templates/design-drafts/WFD-*.json for in-progress designsEdit existing template:
--edit pathRead phases/01-parse.md and execute.
Objective: Extract structured semantic steps + context variables from natural language.
Success: design-session/intent.json written with: steps], variables], task_type, complexity.
Read phases/02-resolve.md and execute.
Objective: Map each intent step to a concrete executor node.
Executor types:
skill — invoke via Skill(skill=..., args=...)cli — invoke via ccw cli -p "..." --tool ... --mode ...command — invoke via Skill(skill="<namespace:command>", args=...)agent — invoke via Agent(subagent_type=..., prompt=...)checkpoint — state save + optional user pauseSuccess: design-session/nodes.json written with resolved executor for each step.
Read phases/03-enrich.md and execute.
Objective: Build DAG edges, auto-inject checkpoints at phase boundaries, validate port compatibility.
Checkpoint injection rules:
skill → skill transition that crosses a semantic phase boundaryagent spawnSuccess: design-session/dag.json with nodes], edges], checkpoints], context_schema{}.
Read phases/04-confirm.md and execute.
Objective: Visualize the pipeline, present to user, incorporate edits.
Display format:
Pipeline: <template-name>
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
N-001 [skill] workflow-lite-plan "{goal}"
|
CP-01 [checkpoint] After Plan auto-continue
|
N-002 [skill] workflow-test-fix "--session N-001"
|
CP-02 [checkpoint] After Tests pause-for-user
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Variables: goal (required)
Checkpoints: 2 (1 auto, 1 pause)AskUserQuestion:
Success: User confirmed pipeline. Final dag.json ready.
Read phases/05-persist.md and execute.
Objective: Assemble final template JSON, write to template library, output summary.
Output:
.workflow/templates/<slug>.json — the reusable templateSuccess: Template saved. User shown: Skill(skill="wf-player", args="<template-path>")
| Spec | Purpose | |------|---------| | specs/node-catalog.md | Available executors, port definitions, arg templates | | specs/template-schema.md | Full JSON template schema |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-19 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +64 percentage points is the difference between those two pass rates over the 18 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.