crewly

Plan and create a NESTED team hierarchy (parent coordination team + per-stream child teams) for a complex multi-stream goal. Use INSTEAD of recommend-team/materialize-team when the goal names 2+ parallel work-streams (e.g. frontend + backend + DevOps). Onboarding-only.

Search the live web for a query and return ranked results (title, URL, snippet) as JSON. Works with no API key (DuckDuckGo HTML fallback) and uses the Brave Search API or SerpAPI automatically when BRAVE_API_KEY / SERPAPI_KEY is set. Any agent can call it for research, competitor analysis, market/trend lookups, fact-checking, finding sources to cite, or grounding a non-software goal (marketing, growth, commerce) in real-world information.

Transcribe a local audio (or video) file to text with timestamps using Whisper. Defaults to local whisper.cpp (free, offline, word/segment-level timestamps) and automatically falls back to the OpenAI Whisper API (whisper-1) when the local engine is not installed. Any agent can call it for meetings, podcasts, voice notes, interviews, or video audio tracks.

Accept and take the next available task from the task queue. Use when an agent is idle and ready to pick up the highest-priority unassigned task. For assigning tasks to specific agents, use assign-task instead.

Register the agent as active with the Crewly backend on startup. Use when an agent first launches and needs to join the team and become visible to the orchestrator. For confirming responsiveness after registration, use heartbeat instead.

Analyze git changes and produce a structured code review with automated checks for missing tests, debug statements, potential secrets, large changes, and dependency modifications. Use when reviewing staged changes, unstaged diffs, recent commits, or branch comparisons before submitting a pull request. For task-level quality gates, use check-quality-gates instead.

Assign a task to a specific agent via the task management system. Use when the orchestrator needs to target a particular agent with a known task ID. For agents self-assigning the next available task, use accept-task instead.

Send a message to all active agent sessions, excluding the orchestrator. Use when the orchestrator needs to announce information or coordinate the entire team at once. For messaging a single agent, use send-message instead.

Record a learning or insight gained during task execution for team knowledge sharing.

Package and submit skills to the Crewly marketplace. Validates, archives, and submits for review to both local and remote registries.

List the recurring cron tasks the orchestrator has scheduled FOR you — call before self-scheduling so you don't create duplicates that fire N× reports.

Schedule a future check-in that creates a WorkItem for you or another agent at a specific time (one-shot or bounded recurring).

Query standard operating procedures relevant to your current context or task.

Author or update a custom SOP (a reusable procedure / playbook) for your team.

Read your team's norms (operating agreements — canDelegate, escalation, rules of engagement) relevant to the current moment.

Propose or update a team norm (an operating agreement — canDelegate, escalation, rules of engagement) for your team.

Decompose a parent Mission (OKR) into child OKRs one cascade tier down (company→team→project) as a PROPOSAL. The runtime drafts the child objectives + Key Results; the skill submits them to the backend as pending_approval. Children are NOT active until the human owner approves. Output includes an [APPROVE] block naming the parent and proposed children for the owner's decision.

Detect + remove low-quality pages from a wiki vault (low confidence, agent memory dumps, per-agent overflow). Default dry-run; archives every deletion; idempotent.

Deterministic vault validation — frozen-path violations, dangling wikilinks, orphan pages, stale claims, restructure proposals. Reports only; the agent decides what to fix.

One-shot conversion from legacy `.crewly/knowledge/*.json` + loose `.md` + agent `memory.json` into the v2.1 three-scope vault structure. Default dry-run; idempotent; never deletes legacy files.

Claim the next pending wiki-queue item and get the vault context your LLM needs to classify it into llm-curated/<folder>/<page>.md. No preset taxonomy — your LLM picks the target.

Read a project/team/global vault and return the system-context payload your runtime feeds to the LLM for synthesis. Single-vault scope (Phase 1).

Vault health report — md counts, duplicate clusters, queue snapshot, consolidation recommendations. Returns signals; your LLM decides what to merge or archive.

Append a source (spec / decision / chat / pr / learning) into a vault's llm-curated/ tree. Refuses writes into frozen paths.

Enqueue wiki-worthy content (from a chat turn, spec write, PR merge, learning) for later agent-driven classification + ingest. Agents must justify with --reason.

Create a WorkItem in the TaskPool targeting a specific agent. The Reconciler will automatically wake the agent and deliver the task when resources are available. No direct agent start or auto-monitoring — the Reconciler handles lifecycle, retries, and resource control.

Proactively notify the orchestrator when a task is done, blocked, or failed.

Create a V3 Request from a user message that the orchestrator has determined to be actionable work.

Perform a system health check and heartbeat. Calls multiple API endpoints to update heartbeat via middleware and returns system status.

Mark a task as complete with a summary of the work done.

Mark a task as complete in the task management system.

Cancel a follow-up trigger previously created with `schedule-followup` or `watch-for-event`, by trigger id or by team-scoped name.

List every follow-up trigger owned by your team — call before creating a new watcher to avoid duplicates, or when auditing why something keeps firing.

Send an agent-authored message back to a chat channel — used by the chat MVP dispatch loop after processing a `[CHAT:<channelId>]` prompt.

Subscribe to a system event (like agent:idle or task:completed) and create a WorkItem every time it fires — reactive follow-up, not time-based.

Convert markdown articles into XiaoHongShu (小红书) style portrait image cards (1080×1440 PNGs) for social distribution.

Score a completed task's quality (0-100) for tracking per-agent quality metrics.

Cancel all active scheduled checks, optionally filtered by session or age (nuclear cleanup).

List all active scheduled checks (one-time and recurring), optionally filtered by session name.

Update a team member's properties — name, role, runtimeType, systemPrompt — via the team membership API.