Trellis Workflow Enhancer
An opt-in Trellis workflow integration skill for AI coding agents. It improves an existing Trellis workflow without handing control to another framework. It inventories the repository, verifies currently installed and upstream Matt Pocock and Waza capabilities, then presents a before-to-after decision table. Nothing changes until the user chooses options; after selection, the integration is applied and verified through the real prompt hook instead of being declared complete from documentation alone.
What It Produces
The first run is analysis only. It gives the user:
- evidence from the repository's Trellis config, workflow, specs, tasks, and real validation commands;
- current local and upstream skill evidence, without silently updating either;
- an explicit enhancement comparison with benefit, cost, risk, prerequisites, and the exact affected files;
- a
no changeoption alongside every recommended bundle.
An applied integration also produces .trellis/skill-integration.json, which
records each skill's invocation mode, Trellis phase, durable artifact handoff,
gate, hook markers, and review order.
Before And After
| Before | Optional after |
|---|---|
| Generic workflow suggestions that may ignore repository shape | Package/spec scope grounded in the real Trellis parser and module roots |
| Planning is shallow when product or domain decisions are unclear | One verified deep decision route, with decisions returned to Trellis artifacts |
| Tests, diagnosis, and manual checks blur together | A named feedback loop, TDD only at a real seam, and reproduction before repair |
| All changes receive the same review cost | Independent review only for declared risk classes |
| Visual work can escape the engineering plan | Waza visual exploration returns states and acceptance criteria to design.md |
| Skills are installed or mentioned but never surface at runtime | Workflow-state routing, a configured prompt hook, artifact contracts, and smoke evidence |
| Tool popularity decides integration | Observed gaps, explicit tradeoffs, and user-selected IDs decide integration |
How It Works
- Inventory: read the target's Trellis files, task state, specs, hooks, code structure, and executable validation surface.
- Verify: inspect installed Matt/Waza skills and compare them with their official upstream sources in read-only mode.
- Compare: show a compact decision table, including
I1runtime binding andN0no change. - Choose: wait for the user to select IDs, bundles, or no change.
- Apply: update only selected project-local workflow routes and write the machine-readable integration contract.
- Prove: run the verifier and prompt-hook smoke test. Report installed, routed, phase-bound, hook-configured, runtime-observed, and integrated as separate evidence levels.
Trellis continues to own task lifecycle, package scope, specification loading, and acceptance evidence. Matt and Waza are never copied into the project or treated as a competing task system.
Install
npx skills@latest add lei1024/trellis-workflow-enhancer
Use
Use $trellis-workflow-enhancer to inspect this repository's Trellis workflow,
compare it with the latest verified Matt and Waza skills, and present optional
enhancements before changing any files.
Optional Bundles
| ID | Bundle | Typical trigger |
|---|---|---|
| I1 | Runtime binding and evidence | Any selected integration; binds routes to state blocks, hooks, artifacts, and smoke verification |
| S1 | Scope and specs | multi-root codebase, stale templates, or ambiguous ownership |
| D1 | Decision and knowledge | unclear product rule, domain term, UX state, or module ownership |
| F1 | Feedback and diagnosis | behavior change, missing confidence, bug, or performance regression |
| R1 | Independent review | shared module, auth, migration, public contract, or high-risk UI path |
| V1 | Visual iteration | UI behavior, hierarchy, responsive state, or screenshot-grounded change |
| H1 | Workflow health | non-trivial repository needs a read-only maintainability audit |
| N0 | No change | existing Trellis workflow already fits the repository |
The catalog and comparison template are intentionally generic. The skill removes unsupported rows rather than forcing every bundle into every repository.
Any applied bundle includes I1. I1 does not make a user-invoked skill run
automatically; it proves whether the hook surfaced the route and whether the
skill result has a Trellis artifact handoff.
Verify A Real Integration
After selecting options, run the bundled verifier against the target repository:
python3 <path-to-trellis-workflow-enhancer>/scripts/verify_integration.py \
<target-repository> --smoke
The command fails when the manifest, local skill source, route table, phase
breadcrumb, hook wiring, invocation mode, artifact handoff, or fixed
trellis-check-before-review order is missing. --smoke executes the configured
per-turn hook once and checks its output markers.
Safety Rules
- Never modify a target before an explicit option selection.
- Never call an integration active because a skill is installed or a route is
written in prose; require
skill-integration.jsonand verifier evidence. - Never claim a locally installed skill is the latest without an upstream check.
- Never auto-install or update Matt, Waza, Trellis, global settings, or hooks.
- Never change
.gitignore, stage, commit, push, archive tasks, or rewrite history as part of an enhancement recommendation. - Never invent test commands. State the missing test or manual-verification gap.
- Never expose secrets, private URLs, user data, or internal infrastructure in the comparison report or durable local documentation.
Sources
The skill uses the official Matt Pocock skills repository and Waza repository as read-only upstream sources. It verifies names and versions at analysis time because both catalogs change independently.
Repository Layout
trellis-workflow-enhancer/
├── SKILL.md
├── README.zh-CN.md
├── agents/openai.yaml
├── assets/readme/hero.svg
├── scripts/verify_integration.py
├── tests/test_verify_integration.py
└── references/
├── comparison-template.md
├── integration-contract.md
└── integration-catalog.md
No comments yet
Be the first to share your take.