engine-control-plane
/engine-control-plane
Use when coordinating or auditing multi-engine workflows, agent roles, command surfaces, hooks, evidence contracts, handoffs, or bounded recovery. Keeps domain doctrine in the owning engines.
- Author
- peterbamuhigire ↗
- Repository
- chwezi-dev-engine ↗
- Licence
- MIT
- Last repository change
- 24.10.2026
Assessment and security
- Files in this package
- 2
- Scripts
- 0
- HTML comments
- 2
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 24.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- SKILL.md zawiera komentarze HTML (2). Nie widać ich w podglądzie Markdown, a model je czyta.
What it does and when to use it
Use when coordinating or auditing multi-engine workflows, agent roles, command surfaces, hooks, evidence contracts, handoffs, or bounded recovery. Keeps domain doctrine in the owning engines.
Original author description
Use when coordinating or auditing multi-engine workflows, agent roles, command surfaces, hooks, evidence contracts, handoffs, or bounded recovery. Keeps domain doctrine in the owning engines.
How it works
- Use When
- Do Not Use When
- Required Inputs
- Prerequisites
- Workflow
- Core content
- Non-negotiables
- Quality Standards
- Anti-Patterns
- Outputs
SKILL.md preview
--- name: engine-control-plane description: Use when coordinating or auditing multi-engine workflows, agent roles, command surfaces, hooks, evidence contracts, handoffs, or bounded recovery. Keeps domain doctrine in the owning engines. metadata: portable: true compatible_with: - claude-code - codex --- # Engine Control Plane Use this skill as the shared operating layer above domain engines. Domain skills remain authoritative for content and technical doctrine; the control plane governs ownership, lifecycle, evidence, handoff, and recovery. <!-- dual-compat-start --> ## Use When - Designing or auditing a workflow that crosses two or more skill engines. - Defining agent roles, thin commands, lifecycle hooks, evidence contracts, or handoffs. - Reviewing release controls, bounded recovery, or a control-plane registry. ## Do Not Use When - A single domain skill can own the work without an orchestration decision. - The task needs domain doctrine rather than routing, ownership, or release control. - A safety-only review is required; load `skill-engine-audit` and its skill safety gate instead. ## Required Inputs - The participating engines, authoritative routers, intended output, and decision owner. - Required capabilities, permissions, dependencies, evidence, stop conditions, and recovery constraints. - The applicable registry or adoption document, if the workflow is already registered. ## Prerequisites - Read the primary domain skill for each participating engine. - Load `skill-writing` and `skill-composition-standards` when changing a reusable contract. ## Workflow 1. **Preflight.** Confirm scope, routers, permissions, dependencies, risk, and the exact mutation boundary. 2. **Assign ownership.** Select one coordinator and the smallest set of independent domain workers. Give each worker a single decision or artefact owner. 3. **Load context.** Carry forward requirements, decisions, constraints, evidence gaps, open risks, and the next owner. Do not reload unrelated doctrine. 4. **Execute thinly.** Use commands and hooks as entrypoints to canonical skills. Keep vendor adapters and command syntax outside model-neutral doctrine. 5. **Check evidence.** Require source, check, result, reviewer, timestamp, and release consequence for each material gate. 6. **Recover or stop.** Preserve the last safe state. Retry only with a named changed approach; escalate after two zero-progress cycles or when authority is missing. 7. **Handoff and learn.** Publish the artefact, evidence, unresolved risks, recovery path, and next owner. Promote a successful repeated lesson only after an independent check. ## Core content ### Required control surfaces | Surface | Minimum contract | |---|---| | Router | One authoritative router and a discoverable catalogue | | Agent | Role, inputs, outputs, permission boundary, and stop condition | | Command | Thin entrypoint to a canonical skill; no duplicated doctrine | | Hook | Event, action, failure mode, adapter, and evidence | | Handoff | Context, decisions, artefacts, open risks, and next owner | | Evidence | Source, check, result, reviewer, timestamp, and release consequence | | Runtime catalog | Exact skill roots, entry count, metadata budget, duplicate check, and host result | ### Agent topology Use one coordinator. Assign independent decisions to a Router/Planner, Domain Worker, Evidence Collector, Adversarial Reviewer, Gatekeeper/Release Captain, or Librarian as needed. Never assign two agents the same decision; the coordinator carries decisions, constraints, risks, and artefact paths forward. ### Hook semantics Implement the same contract through native hooks, scripts, CI, or explicit steps. The minimum lifecycle is `preflight`, `context`, `before_write`, `after_write`, `release`, and `stop`. Advisory telemetry may fail open; safety, evidence, destructive-action, and release gates fail closed or return `NOT ASSESSED`. Two further properties apply to advisory hooks only (safety gates keep failing closed on every attempt): an advisory hook blocks at most once per session, then lets the agent proceed with its note on record; and it softens to a nudge (a message, not a block) when its index or cache is stale for the target file. (Adapted from Graphify-Labs/graphify, Apache-2.0, commit `d6eaa8aae8df155874ebb1044302c055c286342a`.) ## Non-negotiables - Keep domain rules in their owning engine; do not create a second source of truth. - Treat missing tools, sources, screenshots, tests, or approvals as `NOT ASSESSED`, never as `PASS`. - Default review and analysis to read-only. Require explicit authority for mutation or external side effects. - Make stop conditions, rollback or recovery, and release consequences visible before work advances. - Check the assembled runtime skill catalog before release; mark unavailable host or plugin evidence `NOT ASSESSED`. ## Quality Standards - A router is authoritative, discoverable, and checked against the registered catalogue. - An agent has a role, inputs, outputs, permission boundary, and stop condition. - A command is a thin entrypoint and does not duplicate domain doctrine. - A hook names its event, action, failure mode, adapter, and retained evidence. - A handoff names context, decisions, artefacts, open risks, and the next owner. - A runtime catalog gate records exact roots, limits, counts, duplicate names, result, timestamp, and release consequence. ## Anti-Patterns - Two agents own the same decision. Fix: assign one accountable owner and make the other a reviewer. - A command copies a specialist workflow. Fix: route to the canonical skill and keep the command thin. - A missing approval is treated as an implicit approval. Fix: stop and record `NOT ASSESSED`. - A retry repeats the same failed approach. Fix: preserve the prior result and name the changed approach. - A handoff contains prose but no evidence or next owner. Fix: use the evidence and handoff fields as a release gate. - A hook mutates a domain it does not own. Fix: require an explicit cross-engine handoff. ## Outputs | Artifact | Consumed by | Template | |---|---|---| | Ownership and routing plan | Coordinator and domain owners | `docs/engine-control-plane.json` plus inline rationale | | Evidence record | Reviewer or gatekeeper | Evidence contract in `docs/engine-control-plane.md` | | Handoff record | Next owner or release captain | Inline: context, decisions, artefacts, risks, recovery | | Release verdict | Maintainer and operator | `PASS`, `FAIL`, or `NOT ASSESSED` with consequence | ## Evidence Produced | Category | Artifact | Format | Example | |---|---|---|---| | Correctness | Registry and route validation result | Command output | `python scripts/validate_engine_control_plane.py` | | Security | Permission and mutation-boundary record | Markdown | Preflight and before-write evidence | | Operability | Recovery or stop handoff | Markdown | Last safe state and next owner | | Release evidence | Gate verdict with residual risk | Markdown | `PASS`, `FAIL`, or `NOT ASSESSED` | ## Read next - `skill-engine-audit` for whole-engine inventory and scoring. - `skill-engine-audit` (skill safety gate) for unsafe instructions, dependencies, and provenance. - `advanced-testing-strategy` for risk-based test depth and retained evidence. - `world-class-engineering` for implementation and release gates. ## References - `docs/engine-control-plane.md` - shared vocabulary and lifecycle contract. - `docs/engine-control-plane.json` - machine-readable eleven-engine registry. - `../../../scripts/validate_engine_control_plane.py` - deterministic registry validator. - `skill-writing` and `skill-composition-standards` - authoring and composition rules. - [Pointer-rich resumable handoffs](references/pointer-rich-handoffs.md) <!-- dual-compat-end --> ## Inputs | Artefact | Produced by | Required? | Why | |---|---|---|---| | Participating engines and authoritative routers | Task owner or coordinator | yes | Establishes ownership and routing | | Required capabilities, permissions, and recovery constraints | Preflight or risk owner | yes | Bounds execution and side effects | | Registered control-plane contract | Maintainer | conditional | Keeps registry and adoption evidence aligned | ## Capability contract Read and search are required. Editing and execution require an authorised implementation scope. Network access is optional and must remain source- disciplined. Delegation is optional and must use disjoint ownership. ## Degraded mode If a router, registry, tool, source, approval, or test is unavailable, return a scoped plan with the unavailable check marked `NOT ASSESSED`. Do not infer success from a declaration, model confidence, or an unexecuted hook. ## Decision rules | Condition | Action | Failure avoided | |---|---|---| | One engine owns the decision | Route directly and keep orchestration out | Duplicate doctrine and unnecessary hops | | Two or more engines contribute independent artefacts | Use one coordinator, named owners, and a reconciled handoff | Conflicting edits and lost context | | Mutation, external side effect, or irreversible action | Require explicit authority, exact target, recovery, and evidence before execution | Unbounded blast radius | | Required evidence or approval is unavailable | Stop or issue `NOT ASSESSED` with an owner and next check | False release confidence | | A failure repeats with no changed approach | Escalate or defer after the bounded retry budget | Retry loops and hidden failure |
Files in this package (2)
- SKILL.md 9 KB
- references/pointer-rich-handoffs.md 1 KB
Versions
The database has no other skills named engine-control-plane.
Database snapshot: 13.11.2026 · package version fb85cb36e8bc (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100control-plane
Use for any multi-agent dispatch, rotation, concurrency, GitHub workflow, or learning decision. Compact trigger index; load only the bundle the task names. - 100multi-agent-orch
Plans and orchestrates multi-agent swarms, defining roles, topologies, and handoffs. Use when: orchestrate, swarm, delegation, multi-agent, agent-team. - 100oh-my-claudecode
Orchestrate multi-agent Claude Code teams — assign roles, coordinate parallel tasks, share context between agents, and manage team workflows. Use when: running multiple Claude Code agents simultaneously, building team-based AI development workflows, coordinating complex multi-agent coding projects. - 100multi-agent-orchestration
Design and operate bounded multi-agent workflows with task decomposition, dependency graphs, ownership, handoff contracts, shared-state controls, approvals, recovery, and synthesis. Use when a task contains genuinely independent workstreams, specialized roles, parallel research… - 100agent-orchestration
Coordinate superRA agents and handoffs. Requires superRA:using-superra. Use to dispatch roles, run parallel agents, choose reviewers, or preserve workflow state. - 90multi-agent-collaboration
Coordinate multiple agents on implementation or research with isolated ownership, reproducible handoffs, independent checks, and evidence-based conflict resolution. Use when parallel or peer-agent work is requested or already in progress.