team-design

Team Design — Where Are We Going?

Run the DESIGN phase. The design-author decides the approach — recording every self-resolved choice as an auditable assumption — and the adversarial design review gates advancement. No mid-run prompt fires.

Input

$ARGUMENTS is the artifact directory: docs/plans/<id>/. If empty, the discovery command below resolves it.

The design-author reads:

• $ARGUMENTS/1-task.md — what we are building (intent)
• $ARGUMENTS/2-questions.md — the questions that drove research
• $ARGUMENTS/5-research.md — what exists (facts)

Resolve <team-skill-dir> to the absolute directory containing skills/team/SKILL.md. From the repository root, run:

"<team-skill-dir>/discover-topic.sh" "${ARGUMENTS:-}" "5-research.md"

• If the command printed a path, use it as $ARGUMENTS for the rest of this skill (tier 1 explicit arg, or tier 2 discovery). When the path came from tier 2 (no explicit arg), announce the resolved directory to the user before proceeding, so an auto-picked topic is never silent.
• If the command printed nothing (tier 3 — no directory holds 5-research.md), do not hard-error. Fire AskUserQuestion with a Setup header and labeled options:
  • Run the producer — run /team-research docs/plans/<id>/ to produce the missing 5-research.md.
  • Give a path — the user supplies the docs/plans/<id>/ directory directly (run ls docs/plans/ to find your topic directory).

Execution

1. Use the directory resolved in ## Input.

2. Dispatch design-author, which: a. Resolves its own open questions autonomously, recording each in ## Decisions made marked as an assumption (see the agent file) b. Writes $ARGUMENTS/6-design.md with frontmatter revision: 0

   If $ARGUMENTS/6-design.md already exists, skip this dispatch and resume at step 3 — never re-draft an existing design. Both this skip and step 3's never-re-review skip are idempotent re-runs: converge on the same end state, never duplicate work (principle-idempotent-reruns).

3. Design review gate. If the latest $ARGUMENTS/design-review-<n>.md already carries a passing verdict (APPROVE or COMMENT), skip straight to step 4 — never re-review a passed design. Otherwise, before each review dispatch, run the external cross-model pass: call the Skill tool with cross-model-review and follow its ## Design-review pass — reference that procedure, never duplicate it here. Its one gate: the TEAM_DISABLE_CROSS_MODEL kill-switch. Run the runner's detect verb, then run per ready CLI — each through its own named courier sub-agent per that skill's vendor-courier block, with its inline fallback — naming any unavailable CLI to the user per that skill's ## When a vendor CLI is unavailable; a missing runner is skip: cross-model runner not found per CLI. Fence each CLI's raw output as a DATA block at capture time (fence longer than any backtick run in the output, per that section), append one ## External review input section — opening with the untrusted-content line that section specifies — holding the fenced blocks to the review brief, and append the round's transcript to $ARGUMENTS/cross-model-raw.md in the result-line format that section pins (created on first use; a zero-call round appends nothing). Any skip continues with the reviewer alone — the pass never blocks the gate. Then dispatch the adversarial design review (the ## Review brief — call the Skill tool with reviewing-designs to read it, with the artifact directory substituted — run by a fresh-context read-only Explore subagent each round) and write the findings + verdict to $ARGUMENTS/design-review-<n>.md, where <n> is the highest existing <n> + 1 (1 when none exists) — never overwrite an earlier verdict record. Derive the verdict: frontmatter from the last verdict token in the report body — the reviewer's verdict is the terminal line of its report. When the report contains a ### Cross-model disposition section, append that section as one block to $ARGUMENTS/cross-model-notes.md, blockquote-wrapped — prefix every line with > at append time, per the design-review gate in skills/team/SKILL.md — opening with the orchestrator-authored label line — the literal > **Design round <n>** — prepended inside the wrap; same frontmatter-on-first-append rules as the other gates (schema in skills/artifact-frontmatter/SKILL.md). Then act on the verdict:

   • APPROVE or COMMENT — the review passes. Advance.
   • REQUEST CHANGES — re-dispatch design-author with the reviewer's findings verbatim. The agent re-drafts and increments revision: <n+1>, then a fresh review round runs. The loop ends on the verdict, so REQUEST CHANGES keeps re-drafting for as many rounds as it takes. Recovery runs after an operator stop, a context-exhausted session, or the fail-closed halt below. A person revises $ARGUMENTS/6-design.md by hand and re-invokes /team-design bare. The run then resumes at this gate, per the resume branch at step 2. The revision counter persists in 6-design.md frontmatter.
   • Unparseable verdict or reviewer crash — retry the review once with the error; on second failure, halt loudly. Fail closed — never advance on a missing verdict. A missing verdict counts as not passed (principle-fail-closed).

4. Stop once $ARGUMENTS/6-design.md exists and the latest $ARGUMENTS/design-review-<n>.md verdict is APPROVE or COMMENT.