From quantum-loop
Runs quantum-loop execution phase: reads quantum.json, implements eligible stories via TDD/parallel git worktrees, two-stage reviews, and commits until complete or blocked.
npx claudepluginhub andyzengmath/quantum-loop --plugin quantum-loopThis skill uses the workspace's default tool permissions.
Run the autonomous execution loop to implement all stories in quantum.json.
Converts PRDs into machine-readable quantum.json with dependency DAGs, granular 2-5 minute tasks, and execution metadata for quantum-loop autonomous dev pipeline. Use after spec; triggers on 'create plan', 'ql-plan'.
Executes implementation plans using parallel TDD workers for autonomous task completion. Triggers on 'run the plan' or invoke via /plan-orchestrate {plan-name} after plan-create.
Orchestrates 4-phase execution loop (IMPLEMENT, VALIDATE, ADVERSARIAL REVIEW, COMMIT) for complex work units with specs. Verifies outputs adversarially in multi-agent setups.
Share bugs, ideas, or general feedback.
Run the autonomous execution loop to implement all stories in quantum.json.
Before starting:
quantum.json must exist (created by /quantum-loop:plan)If prerequisites are not met, inform the user and stop.
Read and follow the orchestrator agent instructions in agents/orchestrator.md.
The orchestrator will:
Default-on for unattended /ql-execute runs. v0.6.7 + v0.6.8 + v0.6.9 all saw the orchestrator subagent abandon its cycle mid-execution (LLM context-drift). v0.6.8 N6 shipped a prose-only Self-monitoring guard. v0.6.9 N6-followup shipped lib/orchestrator-liveness.sh::poll_orchestrator_commits as a callable runtime helper. This v0.7.0 N14 SKILL-level wrapping is the third and final layer — auto-invoke the helper after dispatching the orchestrator, hand off to the parent on STALE.
# After dispatching the orchestrator subagent (Step 3 of Execution above):
source lib/orchestrator-liveness.sh
wrap_orchestrator_dispatch 600 60 || exit 1
The wrap_orchestrator_dispatch function (v0.7.1 N20 extraction; see
lib/orchestrator-liveness.sh) handles the QL_LIVENESS_ENABLE env-var
check, the poll_orchestrator_commits invocation, and the structured
handoff message internally. The SKILL just calls it and exits 1 on
STALE so CI / wrapper scripts can distinguish stale-signal exits from
clean COMPLETE / BLOCKED exits.
Env var contract:
QL_LIVENESS_ENABLE= | Effect |
|---|---|
(unset) or true | default — wrap orchestrator dispatch with poll_orchestrator_commits (timeout 600s, interval 60s) |
false | skip the wrapping; preserve v0.6.9 dispatch semantics exactly (backwards compat for operators with their own wrappers) |
| any other value | treated as true (default-true semantics; warn to stderr) |
Handoff message structure (printed to stdout when STALE fires):
[QL-EXECUTE] orchestrator-stale signal. header.references/orchestrator-takeover.md (the v0.6.9 N13 SOP).The parent agent reads the handoff and either re-spawns /ql-execute (after confirming whether drift was a transient LLM hiccup) or takes over manually per the SOP. SKILL exits with rc=1 (orchestrator-stale) so CI / wrapper scripts can distinguish stale-signal exits from clean COMPLETE / BLOCKED exits.
Testability note (presence-only AC): the wrapping prose above is consumed by an LLM that runs /ql-execute. There is no runtime test that EXERCISES the env-var conditional or the handoff message structure inside the SKILL itself — those would require running the SKILL in a controlled fixture. Instead, tests/test_ql_execute_liveness_wrapping.sh verifies the prose contains the expected idioms (header, env var, default-true, cross-link, handoff structure). Matches v0.6.8 N6 prose-guard pattern.
For unattended execution outside Claude Code (Linux/Mac):
./quantum-loop.sh --max-iterations 20
./quantum-loop.sh --parallel --max-parallel 4
On Windows, use /ql-execute instead of the shell script for reliable execution.
| Signal | Meaning |
|---|---|
<quantum>COMPLETE</quantum> | All stories passed |
<quantum>BLOCKED</quantum> | No executable stories remain |
<quantum>STORY_PASSED</quantum> | One story completed (more remain) |
<quantum>STORY_FAILED</quantum> | One story failed (will retry if attempts remain) |
<quantum>WAVE_PASSED</quantum> | All stories in a coordinator wave passed (v0.8.0+ coordinator pattern; see agents/coordinator.md) |
<quantum>WAVE_FAILED</quantum> | One or more wave stories failed (parent decides retry/abort) |
After each story's two-stage review gate PASSES but BEFORE the merge to the feature branch, the orchestrator SHOULD invoke /quantum-loop:ql-deslop on the story's changed files. Opt-out via --no-deslop (record the reason in commit trailer Deslop: skipped | <reason>).
Mandatory safety rails, enforced by lib/deslop.sh:
validate_scope <file> BASE HEAD — rejects any edit to a file outside the story's diff. Prevents "adjacent cleanup" scope creep.take_baseline <before.json> — snapshots test / lint / typecheck exit codes.skills/ql-deslop/SKILL.md (dead-code → duplication → consistency → test-reinforcement).take_baseline <after.json> + compare_baseline before.json after.json. Any regression triggers:rollback_pass BASE <files> and emits <quantum>DESLOP_ROLLED_BACK</quantum>. The next pass does NOT start until the user inspects.quantum.deslop[<story-id>].pass_<n>.Language detection: bash lib/deslop.sh detect-language dispatches to the appropriate dead-code detector (knip / ts-prune / vulture / staticcheck / cargo-udeps). When tooling is missing, the step is skipped cleanly rather than failing the pipeline.
After COMPLETE fires, the orchestrator SHOULD invoke /quantum-loop:ql-deep-review before handing the merged branch back to the user. Opt-in via --deep-review (or mandatory when the feature's risk tier is HIGH / CRITICAL per lib/deep-review.sh compute_risk_score).
Gate steps:
BASE_SHA..HEAD_SHA):
bash lib/wave-boundary.sh scan BASE HEAD emits JSON findings for literals that canonicalize to the same key across ≥2 stories (e.g. 'google' vs 'google-api-key'; the Math-Research regression class). Gate via bash lib/wave-boundary.sh gate BASE HEAD — exit 0 = divergence found, exit 1 = clean. Non-empty findings are treated as a medium (2 variants) or high (3+ variants) severity input to the reviewer-set dispatch.lib/resilience.sh run_typecheck).lib/barrel-regen.sh, lib/dep-manifest.sh).bash lib/deep-review.sh score BASE HEAD → then bash lib/deep-review.sh tier <score>.skills/ql-deep-review/SKILL.md §"Risk scoring".lib/deep-review.sh actionability / dedup / hallucination / verdict in sequence to aggregate findings.quantum.reviews[<feature-id>].deepReview.BLOCKS_MERGE → refuse handoff; emit STORY_FAILED with blockers in the failureLog.REQUEST_CHANGES → route to a targeted fix-story in the next wave.APPROVE_WITH_COMMENTS → proceed but log comments into codebasePatterns.APPROVE → proceed cleanly.For legacy pipelines that predate this hook, /quantum-loop:ql-deep-review can be invoked manually post-merge.