| name | orchestrate |
| description | End-to-end orchestrated execution. Takes a feature description and drives the entire pipeline — plan, plan review (3 fresh reviewers), WU decomposition, 4-phase execution loop per WU, final review, COMMIT-READY emit. Honors user's git policy (no auto-commit). |
| source_type | command |
| source_file | commands/orchestrate.md |
orchestrate
Migrated from commands/orchestrate.md.
Codex packaging notes
- Claude/Telar source files remain the source of truth; this file is the generated Codex adapter.
- Skill-local support files from the original Telar skill, such as
references/... or workflow/..., are packaged beside this SKILL.md.
- Repo-root references from the original Telar file, such as
agents/..., commands/..., scripts/..., resources/..., rules/..., hooks/..., or templates/..., are packaged at this plugin root.
- The original Telar orchestration source (
skills/orchestration/...) is packaged under source/skills/orchestration/... for exact-reference lookups; all other Telar skills exist here only as the generated adapters under the plugin-root skills/ directory.
- Resolve plugin-root paths from this generated skill directory via
../.. when reading support files or running packaged scripts.
- In Codex, this skill is the replacement for the Claude slash command
/tl-telar:orchestrate; invoke it as $orchestrate or through @tl-telar.
- Do not require Claude slash-command dispatch or Claude-only environment setup before following the workflow.
- When the original command says to load
skills/orchestration/<name>, load the generated Codex skill at ../orchestration-<name>/SKILL.md first. The original source copy also exists under ../../source/skills/orchestration/<name>/SKILL.md for exact Telar-source references.
- Codex subagent gate — probe, then use or degrade (fail-closed; never fake). Claude
Task() calls map to Codex subagent spawns. Before EVERY multi-reviewer gate: (1) PROBE whether the current Codex surface exposes an agent-spawn tool. (2) If YES → spawn the resolver-selected reviewers as fresh, parallel Codex agent roles; preserve each role, its own rubric, and the freshness rule (no reviewer sees another's verdict or a prior iteration), then close each subagent handle before the next iteration so long runs do not exhaust the local subagent thread limit. (3) If NO → emit a literal DEGRADED: full multi-reviewer gate unavailable on this Codex surface line and STOP the gate. Recommend re-running on a Claude Code host or a Codex build that exposes subagent spawning. NEVER substitute a single inline self-review for the independent multi-reviewer gate, and never silently continue as if the gate passed.
- Stack-aware roster (parity with the Claude path). Derive the reviewer roster from
scripts/tl-telar-reviewer-roster.js (packaged at this plugin root) against the WU file_scope — do NOT hardcode a mobile roster. It returns the domain-correct Security/BackendCorrectness/FrontendUX/Accessibility/Performance reviewers, each with its own rubric path, for mobile, web, backend-data, and rust changes alike.
- Treat Claude
Workflow tool references as unavailable in Codex unless an explicit equivalent tool is present. Use the documented prose fallback path by default.
- Treat
TL_TELAR_ORCHESTRATED=1 as a workflow mode marker in Codex. Do not require a literal Claude slash command to set it.
- Do not pass scheduler
--isolate merely because Codex is running. Use --isolate only after a concrete Codex worktree isolation and merge-back mechanism has been verified for the run; otherwise keep disjoint file-scope serialization.
/tl-telar:orchestrate
Runs the orchestrator playbook in this (main) session. Sets the orchestrated-mode trigger (TL_TELAR_ORCHESTRATED=1) so all orchestrated-namespace skills activate per master design §1.1.
Execution context (binding — do NOT spawn the orchestrator as a subagent)
The orchestrator is a conductor: its job is to spawn fresh Task() subagents — 3 plan reviewers, 6 design reviewers, per-WU implementers, and 2–4 per-WU adversarial reviewers (master design §2.6 Phase α: "All reviewers, implementers spawned as Task()"). A Claude Code subagent cannot use the Task tool — subagents cannot spawn subagents. Therefore the conductor MUST run at the top level.
So this command does NOT call Task(subagent_type=orchestrator). Instead, the main session itself adopts the orchestrator role: read agents/orchestrator.md and execute that playbook directly, in this session, keeping Task available for the reviewer/implementer spawns. Spawning the orchestrator as a subagent silently degrades every multi-reviewer gate to a single inline pass (destroying reviewer independence) and makes per-WU execution impossible.
Behavior
- Set orchestrated mode (
TL_TELAR_ORCHESTRATED=1).
- Capture the user's input verbatim — either the free-text task description, or the
--epic <path> / --plan-file <path> argument (see "Input modes" below).
- Load
agents/orchestrator.md and follow it as the main-session conductor (do NOT spawn it as a subagent). Acting as the orchestrator, you then:
- Run the boot probe (
.tl-telar/ skeleton, safe-default thresholds).
- Wait for / produce an implementation plan — UNLESS
--epic/--plan-file was given, in which case that file IS the plan (drafting is skipped).
- Invoke
skills/orchestration/plan-review-gate (sub-spec 1 deliverable) — spawning 3 fresh reviewer Task()s from this session.
- On PASS: decompose into WUs, write
.tl-telar/plans/active-plan.md, wait for user approval.
- Drive WUs through
skills/orchestration/orchestrated-execution (4-phase loop) using the continuous-frontier dispatch loop (see agents/orchestrator.md → "WU execution — continuous-frontier dispatch"): scripts/tl-telar-wu-scheduler.js computes which WUs are ready (deps COMPLETE + file_scope disjoint from running WUs), bounded by execution.max_parallel_wus (default 3). Up to that many WUs run as concurrent background Task()s from this session; the frontier is recomputed on each WU completion.
- Emit a COMMIT-READY signal for each WU (DOES NOT commit on user's behalf).
- Final review and "Ready for PR" summary.
- Spec Layer archive (conditional). Run
node scripts/tl-telar-spec-archive.js <change-id> ONLY if this run actually produced a Spec Layer change — i.e. a tl-telar-spec/changes/<change-id>/ with a REQUIREMENTS.delta.md exists (created when requirements were gathered via skills/requirements-gather.md → "Step 0"). It merges the delta(s) into tl-telar-spec/truth/ and moves the change folder to tl-telar-spec/changes/archive/<date>-<id>/. If no such change dir exists, SKIP this step — do not run the archive with an unbound/nonexistent <change-id> (it would abort with "change directory not found"). Note: the orchestrator's own working artifacts live under .tl-telar/plans/active-plan.md, which is NOT a Spec Layer change; automatic creation of a tl-telar-spec/changes/<id>/ from within the orchestrator playbook is not yet wired (deferred — see docs/superpowers/specs/2026-07-02-telar-spec-layer-design.md). Today the Spec Layer is exercised through the legacy /tl-telar:add-feature / /tl-telar:create-app / /tl-telar:update-requirement commands, which invoke Step 0.
Input modes
The orchestrator accepts the work to do in one of three forms. All three converge on the same pipeline at the Plan Review Gate (Step 4) — the only difference is how the plan gets produced.
| Mode | Invocation | Plan drafting (Step 3) | Best for |
|---|
| Free-text | /tl-telar:orchestrate <description> | Orchestrator drives/produces a PLAN.md, then reviews it | Quick start; you don't have a plan yet |
| Epic file | /tl-telar:orchestrate --epic <path> | Skipped — the epic file IS the plan; its ## Task sections decompose 1:1 into WUs | A prepared single-feature epic (authored from templates/epic.md) |
| Plan file | /tl-telar:orchestrate --plan-file <path> | Skipped — the PLAN.md is reviewed and decomposed directly | You already maintain a WU-decomposable PLAN.md |
Epic / plan-file rules:
- One feature per file. An epic spanning several independent features is not one plan — split it and run the command per file.
- The file still goes through the mandatory 3-reviewer Plan Review Gate.
--epic/--plan-file skips drafting, never review.
- Author epics from
templates/epic.md — its task fields mirror the Work Unit schema, so decomposition is mechanical.
- If the file's
## Intent (epic) or goal (plan) is missing, the Scope-Alignment reviewer will stop and ask for the original request.
Autonomy (autonomy.cycle in .tl-telar-thresholds.json)
interactive (default) — the run may pause at checkpoint: true WUs and at self-reflect.unattended — exactly ONE human gate: the plan-ready approval (Step 5), which first clarifies UI (ASCII drafts signed off) and collects all secrets/inputs. After "go", the WU cycle runs to PR-ready with no pauses. UI/visual sign-off is hoisted to plan-readiness, never mid-cycle; an uncollected decision STOPs as a pre-flight defect rather than being guessed. See agents/orchestrator.md → "Autonomy model".
Usage examples
/tl-telar:orchestrate Add a login screen with email + password
/tl-telar:orchestrate Wire up Stripe subscription billing for the Pro tier
/tl-telar:orchestrate Refactor the bottom tab navigator to use Material 3 conventions
/tl-telar:orchestrate --epic docs/epics/user-profile.md
/tl-telar:orchestrate --plan-file docs/plans/checkout-flow.md
What this command does NOT do
- Does NOT git add or git commit. User policy is "ben yapacagim" (user handles git manually). The orchestrator emits COMMIT-READY signals with suggested messages.
- Does NOT skip plan review even if the user says "this plan is obviously fine." The 3-reviewer gate is mandatory.
- Does NOT loop past 3 retries on any gate. Escalates to user with structured options.
- Does NOT change legacy commands' review/build behavior —
/tl-telar:add-feature, /tl-telar:create-app, and /tl-telar:update-requirement still don't route through this orchestrator's plan/design gates. They DO now share the same Spec Layer artifact location (tl-telar-spec/changes/<id>/, tl-telar-spec/truth/<domain>/) as this command — see docs/superpowers/specs/2026-07-02-telar-spec-layer-design.md. (/tl-telar:migrate-app is intentionally NOT included — it doesn't invoke requirements-gather or produce a REQUIREMENTS.md.)
Comparison with legacy commands
| Legacy | Orchestrated equivalent |
|---|
/tl-telar:add-feature <feature> | /tl-telar:orchestrate <feature> |
/tl-telar:create-app <description> | /tl-telar:orchestrate Create a new app: <description> (treats whole-app as a single big plan) |
| Manual review at each phase | Automated 3-reviewer gate at plan + a risk-tier-scaled per-WU review gate (trivial→1, standard→~2, critical→full roster) + always-on cross-model second review when configured |
Choose orchestrated mode when you want rigorous adversarial review at every gate. Choose legacy when the task is small or the existing flow's UX suits you better.
Tests / conformance
Sub-spec 2 ships with no scripted end-to-end test of this command (markdown-instruction tooling has no test runner). Verify manually by running on tests/fixtures/plan-review/sample-good-plan.md:
/tl-telar:orchestrate "Add the CHANGELOG validator from tests/fixtures/plan-review/sample-good-plan.md"
Expected: plan-review-gate fires, PASSes, orchestrator decomposes into the 3 WUs (validator file, CLAUDE.md row, smoke test), drives each through the loop, emits COMMIT-READY signals.
