build-tdds

/build-tdds

Interactive parent skill. It sequences workers; it is not a detached coprocess. Session survival is whatever the harness does.

1. Source helpers (fail closed)

Resolve the plugin tree from the first set of CLAUDE_PLUGIN_ROOT / GROK_PLUGIN_ROOT, then source helpers. If any source fails, stop.

_tl_src="${CLAUDE_PLUGIN_ROOT:-${GROK_PLUGIN_ROOT:-}}"
. "${_tl_src}/scripts/lib/plugin-root.sh" || { echo "throughline: cannot source plugin-root.sh" >&2; exit 1; }
. "$(tl_plugin_root)/scripts/lib/verdicts.sh" || exit 1
. "$(tl_plugin_root)/scripts/lib/run-record.sh" || exit 1
. "$(tl_plugin_root)/scripts/lib/models.sh" || exit 1
REPO="$(git rev-parse --show-toplevel)"

REPO is the human session / integration checkout. Never pass the build worktree path as . Every tl_verdict_* and tl_run_* call uses "$REPO".

2. Lock (FR-18 / FR-43)

tl_run_lock "$REPO". If held by a live PID → refuse. If held by a dead PID → tl_run_lock_reclaim "$REPO".

3. Resume vs fresh (FR-39 / FR-40)

If docs/tdd/.implement-logs/latest has a TDD whose status is not terminal (done|failed|blocked|skipped):

Ask the user a structured question: Resume / Start fresh.

• Fresh: delete latest state fragments only (run.json and <slug>.json). Keep logs. Then continue at step 4.
• Resume: g="$(tl_run_next_gate "$REPO" "$run" "$slug")". Jump:
  • test-first — if the build branch exists and has commits beyond integration, do not re-dispatch the implementer; run step 8 observe-only. If the branch has no commits beyond integration, dispatch the implementer (step 7) with "continue from existing branch".
  • ci-checks — skip 7–8a; run ci-checks.sh only (step 8b).
  • runtime-verify — skip 7–8; run step 9 only.
  • review — skip 7–9; run step 10 only.
  • flip — skip 7–10; run step 11 only.

Never treat partial feat: commits as build-gate completion. The test-first verdict file is the only build-complete signal.