workflow-engine

A fixed workflow runner for coded workflows, where each step and its order are defined in advance. It handles parallel stages, checks outputs, caching, and rerunning affected steps.

In plain words
What is it for?
Use it to compile workflow definitions, run stages in waves, validate results, reuse cached work, and invalidate downstream work when inputs change.
Why use it?
It makes repeated multi-step agent work predictable and reproducible, while avoiding manual decisions about what runs next.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/data-wise/craft/workflow-engine
Any agent
npx skills add Data-Wise/craft --skill workflow-engine
Clone the repo
git clone --depth 1 https://github.com/Data-Wise/craft

Made for: Claude Code, Codex.

Per session 101 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,105 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00101 $0.02105
Opus 5 $0.00051 $0.01052
Sonnet 5 $0.00020 $0.00421
Haiku 4.5 $0.00010 $0.00211

Measured yesterday against content hash 0f0a853471c8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

workflow-engine scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/orchestration/workflow-engine/SKILL.md · 178 lines

How it starts

The opening of the file, as written. The whole thing — 178 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Workflow Engine

The reusable execution body behind /craft:orch:workflow. The command owns args (--dry-run, --resume, --refine); this skill owns the work.

Unlike drive-engine (improvises "what next" each turn) the control flow here is fixed in the definition — only the count of agents in a parallel stage flexes to upstream data. That determinism is real only because the mechanical steps are delegated to scripts/workflow_parse.py, never improvised. You orchestrate and judge the advisory semantic layer; the Python core decides everything that must be reproducible.

Hard rule — delegate the deterministic mechanics

Never eyeball these. Always shell out to the core:

Mechanic Call Decision it owns
Compile plan python3 scripts/workflow_parse.py <file> wave order + fan-out shape (D1/D3)
Structural gate gate_output(data, schema, stage) pass/fail per agent output (D2 layer 1)
Cache key cache_key(stage_block, resolved_input, role_version) replay vs re-run (D4)
Cascade cascade_invalidate(stages, changed, build_deps(plan)) downstream invalidation (D4)
Fan-out resolve_fanout(over, upstream_outputs) bound items; empty → hard abort (D6)
Semaphore sem_acquire/sem_release/sem_reconcile live-agent ceiling (D5)

Agent Prompt Composition (Lever B — prompt-trim)

Every subagent prompt MUST contain exactly three parts — nothing more:

  1. Spec slice — only the portion of the WORKFLOW definition scoped to THIS agent's files/phase. Do NOT pass the entire WORKFLOW-*.yaml to every agent. Extract and pass only the stage block(s) the agent needs to execute.

  2. Summarized prior outputs — a brief structured summary of earlier stage results (~200 tokens max per stage). Do NOT include full transcripts or raw prior-agent outputs. Derive the summary from the structured return values cached in .craft/workflow-runs/<run-id>/.

  3. Structured-return instruction — end every prompt with exactly:

Read the full file on GitHub · 178 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 178 lines · 101 tokens per session scan A 0f0a853471c8

Subscribe to this mod's changes

workflow-engine is a skill published in the GitHub repository Data-Wise/craft (4 stars, last pushed 16d ago), licensed MIT. It adds 101 tokens to every session and 2,105 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

github-actions-workflow

Scaffolds or audits a GitHub Actions CI/CD workflow for a project. Covers job structure, caching, secrets handling, concurrency groups, environment gates, and reusable workflows via workflowcall. Invoked when the user asks to set up CI, add a GitHub Actions workflow, improve pipeline performance, or share workflow…

soulcodex/agentic · 72 tokens

specflow

Spec-driven development with executable contracts.

Hulupeep/Specflow · 9 tokens

specflow-loop-selector

Selects the correct Specflow loop and forces a concrete run contract before work starts. Use when an agent is asked to start Specflow work, build a ticket, create/refine a PRD, investigate an existing product, run Gate D, choose between spec-build and feature-build, or "go find out about" Specflow loops. Compatible…

Hulupeep/Specflow · 106 tokens

specflow-simulate

Simulates end-to-end usage of a Specflow story across multiple personas and divergent routes to discover gaps, edge cases, and unhandled branches BEFORE the story is built — then proposes each finding as a concrete story addition (new REQ, negative-path AC, Gherkin branch, or journey step). This is the high-value…

Hulupeep/Specflow · 172 tokens

specflow-audit

Audits a Specflow story/ticket for spec-compliance and surgically uplifts it — adding only the missing executable contract sections (SQL/RLS, TypeScript interfaces, invariant codes, Gherkin, acceptance criteria, Definition of Done, data-testid coverage) — then runs a pre-flight gate that refuses to mark the ticket…

Hulupeep/Specflow · 199 tokens

architecture-review

Conducts a comprehensive multi-perspective architecture review using ALL architecture team members. Use when the user requests "Start architecture review", "Full architecture review", "Review architecture for version X.Y.Z", "Conduct comprehensive review", or when they want assessment from multiple perspectives. Do…

codenamev/ai-software-architect · 80 tokens