setup
For a new TeamWorkflow installation, binding refresh, or drift audit after a pack release, inspect and bind the project workflow. Reuse explicit recorded decisions with their sources; ask about missing, conflicting, or changed authority and bindings.
timharris707 · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add timharris707/skills --skill setupAbout this skill
name: setup description: "For a new TeamWorkflow installation, binding refresh, or drift audit after a pack release, inspect and bind the project workflow. Reuse explicit recorded decisions with their sources; ask about missing, conflicting, or changed authority and bindings."
Setup
Read the Codex desktop binding when this workflow needs harness mechanics, model routing, or recovery.
This skill binds the team-workflow pack to a specific repo, once, through an interview, and refreshes that binding on re-run. Everything the other pack skills call "the binding doc" is what this skill seeds. Nothing here rewrites the repo's CI or processes; it records how the pack composes with what is already there.
Posture: inspect first, reuse decisions. Scan the tracker, verification commands, instruction files, templates, installed skill paths, and configuration ownership. Reuse explicit recorded bindings and prior answers under the safeguards below. Ask only about a new binding, material drift, or an unresolved choice. A read-only installation audit needs no setup interview or tracker write.
The five mandatory bindings
-
Tracker. The repo + tracker in use, and the mapping onto the pack's tracker discipline: how claims are posted, how blocking is expressed, what the frontier query is. The pack is GitHub-Issues-first (the recipes in references/tracker-discipline.md use
gh), but the tracker is a named binding, not an assumption baked into the skills: a repo on a different tracker writes a different tracker section in its binding doc, mapping the same discipline (machine-readable claims, dependency edges, a frontier query) onto its own tool. The skills stay unchanged.Implicit-repo check (GitHub trackers): at binding time, compare gh's implicit repo resolution in the working directory (
gh repo view) against the bound repo. A fork or multi-remote clone commonly resolves to the upstream repo, and every recipe run unqualified would then read and write tracker state there, not on the bound tracker. On mismatch, warn and record it in the tracker binding. The recipes scopegh issueandgh labelwith--repo <owner>/<repo>and write the bound repo literally into everygh apipath (see tracker-discipline), precisely so this mismatch stays harmless; do not lean ongh repo set-default, which is per-clone state the next worktree or machine doesn't have.A repo with no tracker yet is a normal starting state, not a setup failure. Offer two honest paths and record whichever the human picks: (a) bind one now: on a GitHub remote that usually means the repo's own GitHub Issues, ready immediately; (b) record
none yetas the tracker binding, with the consequence stated in the doc: the claim, frontier, and blocking recipes stay dormant until a tracker is bound. Never invent a tracker the team doesn't use, and never leave the field blank.Label vocabulary on a fresh tracker: the pack's recipes assume the labels they query already exist: a frontier query against labels nobody created returns empty forever, and nothing downstream ever creates them. When binding a tracker that lacks them, setup enumerates the pack's reference vocabulary (state:
needs-triage,ready-for-agent,ready-for-human,blocked; type:slice,bug,gate-decision,process) and, on confirmation, creates the missing ones (on GitHub:gh label create "<label>" --repo <owner>/<repo>, one per missing label, a declared write) or records the creation instruction in the binding doc for trackers setup cannot drive. A repo with existing labels maps them instead via the optional label-vocabulary binding; nothing gets renamed. -
Verify commands. The exact commands that constitute "verified" in this repo (typecheck, lint, test, build, whatever the repo runs), recorded so every brief and template can name them instead of guessing. A repo with no toolchain yet records an explicit
none yet, never a guessed or aspirational command that would let briefs claim a verification nobody can run. -
The decider. Who adjudicates decisions in this repo. Reuse only an explicit recorded decision-maker binding and cite its file or decision-record source. Ask when that binding is missing, conflicting, or there is evidence it changed; never infer authority from Git authorship, an account, or a remembered name. Current user instructions take precedence over a recorded role. Until resolved, pause only work that depends on that authority. Every pack skill says "the decider"; this binding is where that role gets a name. A repo without a clear answer here is not ready for the pack's decision discipline; surface that honestly.
-
The binding-doc home. Where the anchor binding doc lives. Default:
docs/agents/team-workflow.md. Reuse the confirmed home, and ask only when it is new or changing: repos with config-distribution pipelines can make the default location wrong in ways only a human knows.The home must be git-tracked and outside anything a sync pipeline manages, stated so a config-pipeline owner can sign off from reading alone: (a) bindings are shared, per-repo team decisions, so they belong in version control where edits are reviewed and history survives, not in per-machine synced files; (b) sync pipelines republish and checksum what they manage, so a binding doc inside a synced tree is either clobbered on the next sync or turns every team edit into a fight with the pipeline.
-
Working mode. Ask, in these words or close to them: "Will you read the code your agents write, or lead from outside it?" Asked on first bind and never inferred from the repo; a re-run reuses the recorded answer unless it is missing or the human changes it.
leadrecords that the human works from the orchestrator seat: new sessions in this repo open by invoking orchestrate and following its startup