AI Efficiency Operating System

Plugin ai-efficiency-operating-system in xq22115/ai-thinking-debate-skills-lab. Commands read from the repo's manifest — 11 days ago.

Skills (17)

• adversarial-deliberation-council

  Auto-invoke the repository's canonical adversarial deliberation council for complex, contested, cross-system, research-heavy, or repeatedly failing tasks. Produces evidence-bearing specialist disagreement and runtime-verifiable synthesis without exposing private chain-of-thought.

• agent-runtime-forensics

  Use when an agent's behavior, side effects, provenance, or completion is disputed and model/tool transcripts alone cannot prove which decision, tool call, process, file, network event, or artifact caused the observed result.

• authorized-reverse-engineering

  Use when analyzing software, firmware, binaries, or protocols the user is authorized to inspect and static or dynamic reverse engineering is needed to understand behavior, compatibility, architecture, regressions, formats, or undocumented interfaces.

• autonomy-contract

  Use when the user explicitly asks for autonomous or effect-bearing work, broad delegation, unattended execution, or actions whose approval and irreversible boundaries must be fixed before execution.

• capability-forensics

  Use when an AI model, agent, plugin, desktop workflow, or tool appears unable to do something and the real bottleneck could be the model, harness, tool surface, permission, session, context, entitlement, environment, or runtime state.

• chief-of-staff-core

  Use when a task is complex, multi-stage, multi-tool, long-horizon, or constraint-heavy and needs one primary phase owner to preserve the user's objective, acceptance criteria, authority and routing.

• convergence-controller

  Use when repeated review, iterative repair, many reviewer comments, skill evolution, or keep-improving loops risk scope creep, stagnation, regression or endless retries.

• evidence-watchdog

  Use when a claim of fixed, configured, installed, tested, live, deployed, delivered or complete must be proven, or when a stateful mutation requires postcondition/read-back evidence.

• executive-research

  Use when current, uncertain, conflicting or technical evidence matters; when deeper reasoning, root-cause diagnosis, temporal/version analysis, counterevidence, or broad source triangulation can change the decision.

• github-operation-orchestrator

  Use when GitHub work requires a reliable multi-step read/search/analyze/tool-call/write/PR/workflow/verification loop, especially for plugin or skill pulls, search underfetch, repeated connector failures, partial results, branch mutations, or claims that source changes are actually live.

• mcp-surface-engineering

  Use when an AI agent has a large, changing, expensive, conflicting, user-specific, or untrusted MCP/tool surface and tool discovery, schema drift, context cost, entitlement, namespace collision, or tool-poisoning risk may affect correctness.

• memory-policy

  Use when deciding what should persist across turns or sessions, recovering prior context, reconciling stale or conflicting memory, or handling instruction-like content from external sources.

• persistent-work-ledger

  Use when the user explicitly needs durable cross-session state, resume, concurrency, scheduling, handoff or side-effect reconciliation and the current host actually exposes filesystem or durable-state primitives.

• plan-arbiter

  Use when choosing between implementation routes, architectures, sequencing options, tradeoffs, or recovery plans before effectful execution.

• root-cause-falsification

  Auto-invoke the canonical root-cause-falsification policy for debugging, repair, optimization, flaky behavior, performance issues, capability claims, repeated workaround stacks, or completion claims that can be falsified by a real counterexample.

• superpowers-conversation-runtime

  Use when ordinary ChatGPT work involves debugging, implemented behavior changes, feature implementation, code-review feedback, or creating/editing an agent skill.

• task-goal-intelligence

  Use when a substantive task could be misread because the real target, constraints, corrections, route ownership, hidden requirements, or completion condition can materially change execution.

Similar plugins