agentleFS
mhyllemhylle/claude-skills-collection/skills/tt-workflow-build/SKILL.md

tt-workflow-build

Skill20 starsChanged 36 days agoMITView on GitHub

Tasktracker-native trigger for a PARALLEL build. Thin by design: (1) ensure a tasktracker project exists (use the existing one, or create one), then (2) build it in dependency waves, fanning each wave's independent slices out to worktree agents — through the Claude Code Workflow tool when it is in the tool list, otherwise parallel Agent subagents — while the parent integrates, gates and tracks every slice in tasktracker using the build and verify skills. It does not analyze parallelism up front, ask the user to choose a mode or hand back; it drives autonomously to done, pausing only at real human gates such as deploy. Use when the user wants a build driven by parallel agents, typically after /brainstorm and /tt-create-plan, or standalone. Prefer /tt-implement-plan or /tt-workflow-run only when the whole job is irreducibly sequential; prefer /tt-workflow-audit for read-only analysis.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 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.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If none exists — create one: Prefer running /tt-create-plan (it yields the requirements + phased tasks the build consumes).

    line 46
  2. Distinctive:

    Never reach Step 2 without a project.

    line 48
  3. Very specific:

    Get a project, then start the build — autonomously: Never ask the user to pick a mode, never hand back, never abandon to /tt-implement-plan …

    line 61

≈ “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

tt-workflow-build

<!-- contents: generated by tests/skill_structure.py --write; edit headings, not this block -->

Contents

  • Step 1 — Ensure a project (the one hard precondition)
  • Step 2 — Start the parallel build
  • Non-negotiables
  • References

Reference files (open one when a step points to it):

  • references/build-workflow-protocol.md: Building the build workflow — construction detail
  • references/workflow-tasktracker-contract.md: Workflow ↔ TaskTracker Contract
<!-- /contents -->

A thin trigger, not a hand-rolled orchestrator. When invoked, do exactly two things — then drive to done:

  1. Ensure a tasktracker project exists.
  2. Start the parallel build — dependency waves of worktree agents, through the Workflow tool when you have it and parallel Agent calls when you don't — tracking the work in tasktracker, using the build + verify skills.

Do NOT stall. No "is this parallel yet?", no asking the user sequential-or-parallel, no handing back, no switching to /tt-implement-plan. Invoking this skill is the decision to build in parallel. Get a project → start the build → run it to done autonomously, pausing only at genuine human gates (e.g. deploy).

Step 1 — Ensure a project (the one hard precondition)

A build needs a tasktracker project to build into and track against. We need one, so:

  • If a project already exists — the argument-hint slug/id, the session's active project, or a clear match from tasktracker_listProjects — use it. (Normally /brainstorm → /tt-create-plan already produced a planned project with requirements + a phased, dependency-graphed task tree — that's the ideal input.)
  • If none exists — create one. Prefer running /tt-create-plan (it yields the requirements + phased tasks the build consumes); at minimum tasktracker_createProject.

Never reach Step 2 without a project. That's the only thing this skill guarantees before starting the build.

Step 2 — Start the parallel build

Pick the engine from your tool list (the contract's "Tool availability" section):

  • Workflow is in your tool list — declared to you directly, or among the deferred tools (load it with ToolSearch select:Workflow) → each wave is one Workflow run whose script fans the wave's slices out as worktree agents.
  • It isn't — Workflow is opt-in, so this is the common case → each wave is one message holding one Agent call per slice, with subagent_type: "general-purpose" and isolation: "worktree", the same slice prompt the workflow script would send, and the same SLICE_RESULT asked for as JSON. Wait for every completion notification of the wave before integrating.

Either way the parent does the rest the same way: integrate each returned slice, gate it, track it in tasktracker, select the next wave, loop until done. Read references/build-workflow-protocol.md (and the contract it points to) and follow it — slice schema, slice prompt, wave construction for both engines, integration, gates — but treat it as construction detail, not a decision gate that can stop you.

Non-negotiables

  • Get a project, then start the build — autonomously. Never ask the user to pick a mode, never hand back, never abandon to /tt-implement-plan / /tt-workflow-run.
  • Parallel work runs in parallel, engaged as early as the dependency graph allows — through Workflow when it is in your tool list, otherwise through parallel Agent calls. A missing Workflow tool changes the engine, not the plan. The only sequential work is the irreducible bridge — the one-time git init + scaffold + first commit a fresh repo needs before any worktree can branch, or a lone dependent leaf — done inline by the parent as part of the build (it makes the next parallel wave possible). Never run ready, independent slices one after another.
  • Parent owns every tasktracker write (parallel agents share one MCP identity and cannot write — contract R1); agents return schema results only.
  • The tool list decides, not ToolSearch — a tool declared to you directly never appears in ToolSearch, so a miss there proves nothing.

References

  • references/build-workflow-protocol.md — how the parent constructs the build (slice-result schema, slice prompt, the Workflow and Agent wave, git-apply integration, per-slice gates, PARTIAL/BLOCKED handling, wave loop, resume). Follow it; it is NOT a stall gate.
  • references/workflow-tasktracker-contract.md — the shared tt-workflow-* contract: tool availability (check the tool list; Agent fallback when Workflow is absent), parent-owns-writes, worktree isolation + integration intents (R6), no wall-clock/RNG in scripts, quality gates, prod safety, canonical skeleton. Read it before building.

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.