coordinator-kit

Plugin coordinator-kit from marketplace coordinator-kit in nikolamin/claude-coordinator-kit. Commands read from the repo's manifest — 2 days ago.

Skills (15)

• agent-brief-hygiene

  Write bounded, self-contained coordinator dispatch briefs with original requirements, falsifiable premises, resource ownership, evidence paths, and safe handoff rules. Use before dispatching a builder, investigator, fixer, or verifier.

• backlog-discipline

  Route coordinator follow-ups and external signals into the canonical task record, check stale ticket premises, and preserve the founder's original intent. Use at intake or when an agent discovers work outside its current scope.

• bootstrap

  Initialize or resume a coordinator through the bundled database, automatically migrate legacy STATE/plan/decision companions on first updated-plugin use, reconcile imported records, and restore agent/inbox ownership without replacing project instructions.

• codex-second-opinion

  Set up and invoke the optional Codex CLI for a coordinator agent's independent opinion on UI/UX, copy, research, or document review. Use when that second-opinion path is selected; escalation decides when it is useful.

• comms-register

  Format coordinator status answers and notifications by user goal, with a project-specific quiet or checkpoint policy. Use when reporting outcomes, blockers, or status over chat or an authorized bridge; decision wording belongs to question-protocol.

• coordination-state

  Use the bundled coordinator SQLite database for tasks, events, decisions, questions, lanes, profiles and handoffs. Run first-use migration of STATE/plan/decision companions, reconcile imported records, then read and write canonical state through the CLI.

• credential-handling

  Use task-authorized account access while keeping credentials out of transcripts, arguments, derived artifacts and coordinator state. Use before dispatching work involving login, secrets or database access; restate the operative constraints in the brief.

• escalation

  Escalate repeated coordinator verification failures or unresolved architecture/judgment questions to advice or an optional second model. Use after the bounded retry loop is exhausted or another perspective could resolve a material choice.

• execute-loop

  Run the coordinator's task loop, assign exclusive work lanes, bound retries, and gate pushes and completion on revision-specific evidence. Use when dispatching work, deciding what can run concurrently, or checking whether a task can ship. For how to prove behavior, use verification-standard.

• phase-loop

  Choose between greenfield project phases and continuous coordinator operation, define plan/validation gates, and locate the knowledge base. Use for lifecycle or planning questions; per-task delivery is in execute-loop and initialization is in bootstrap.

• question-protocol

  Present one coordinator decision at a time, correlate terse replies to their exact question, and preserve approval scope. Use for an actual founder-only choice or blocker, not routine authorized work or a status answer.

• stop-and-save

  Honor a coordinator stop request, preserve in-flight work and pending decisions, and write a durable handoff. Use on "stop and save your step" or equivalent; bootstrap handles the subsequent resume.

• ux-audit

  Run a requested pre-launch UX audit with a feature-by-persona matrix, isolated browser tests, and HTML report. If the user has not supplied scope, ask for scope immediately and wait before any project inspection, file listing, bootstrap or agent dispatch. Do not infer scope by scanning the workspace

• verification-standard

  Establish risk-appropriate acceptance evidence, fail-first controls, live UI checks, and honest test coverage. Use when choosing or evaluating verification or before sharing a demo. Push scheduling and evidence reuse belong to execute-loop.

• watchdogs

  Monitor in-flight coordinator work, distinguish slow agents from lost ones, recover interrupted trees, and restore inbox ownership after restart. Use on wakeups, missing completions, or suspected stalls; use bootstrap or stop-and-save for session handoff.

Hooks (9)

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/session-start.sh"

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh" context

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh" agent

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh" bash

• PostToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh" context

• PostToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/autolane.sh" post

• UserPromptSubmit

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/guard.sh" context

• PreCompact

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/checkpoint.sh"

• SubagentStop

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/autolane.sh" stop

Similar plugins