SwarmAI

Orchestrate the full AIDLC Autonomous Pipeline from a one-sentence requirement to a PR-ready delivery with TDD methodology. DDD drives judgment, SDD produces specs, TDD verifies delivery. Stages: Evaluate, Think, Plan, Build, Review, Test, Deliver, Reflect. Checkpoints on BLOCK or context limits. TRIGGER: "run pipeline", "autonomous pipeline", "pipeline for", "full pipeline". NOT FOR: single stages (use evaluate, deep-research, code-review, qa, deliver directly).

DDD-native autonomous dev loop — judge→execute→reflect on a SINGLE DDD via a bundled, decoupled engine (engine/artifact_cli.py, file-based .artifacts/ state, NO data.db, NO SwarmAI backend). Retains the moat: Gate-2 adversarial-before-commit + the 养成 ladder. DDD-native decouple of SwarmAI's s_autonomous-pipeline engine. TRIGGER: "ddd pipeline", "run the ddd loop", "build in this ddd". NOT FOR: SwarmAI's own multi-project pipeline (that's the native s_autonomous-pipeline).

Package a grown DDD into a distributable capability package — the last step of the DDD lifecycle (cultivate → version → DISTRIBUTE). Human-in-the-loop: reads the DDD's OWN declared reach (aim.json distribution block = the ceiling), confirms which declared target(s) to emit NOW, renders BOTH standard shapes on demand — an internal AIM capabilities package (Config+AIMBuild+agent-spec) and/or an Open-Plugins plugin (.plugin/plugin.json) — runs a fail-closed content-safety scan over the emitted tree, and returns the built package path. Never widens reach by inference; emit ≠ publish. TRIGGER: "distribute this ddd", "package ddd", "ship ddd to kiro/quick/claude code", "publish ddd", "make a plugin from this ddd". NOT FOR: creating/binding a DDD (s_ddd-manager); running the pipeline (s_ddd-pipeline); publishing content/media (s_ddd-pollinate).

Create, list, edit, rename, and delete DDD projects (Domain-Driven Design) — provisions the canonical six-section DDD structure. Each project gets 4 knowledge documents (PRODUCT.md, TECH.md, IMPROVEMENT.md, PROJECT.md) and an .artifacts/ directory for pipeline outputs. TRIGGER: "add project", "new project", "create project", "rename project", "create DDD". NOT FOR: radar-todo use cases.

Estimate token count for files using the bundled script. TRIGGER: "estimate tokens", "count tokens", "token usage", "context window size". DO NOT USE: for manual token math — always run the script.

Full SwarmAI release cycle: preflight checks, version bump, binary build, desktop package, smoke test, and GitHub publish. Supersedes s_release (which only does version bump + tag) by adding build/package/smoke stages. TRIGGER: "release", "cut release", "ship it", "发版". NOT FOR: s_swarm-build, s_release, s_swarm-hive use cases.

Persist knowledge to the correct destination — routes to DDD docs (PRODUCT/TECH/IMPROVEMENT/PROJECT), MEMORY.md, Knowledge/Library/ (searchable store), or EVOLUTION.md based on content type. Unified routing for both manual saves and auto hooks. TRIGGER: "remember", "save", "persist", "沉淀", "record this", "save to DDD", "save lessons". NOT FOR: save-context, save-activity, self-evolution (governance rules) use cases.

Generate AI-Ready artifacts for any codebase — transforms repos into DDD-structured context (AGENTS.md + PRODUCT/TECH/IMPROVEMENT/PROJECT.md + code-intel.json) that makes AI agents truly understand the project. Works on any language, any framework. TRIGGER: "make AI-ready", "ai-ready repo", "generate DDD for repo", "analyze codebase". NOT FOR: code-review, deep-research use cases.

The ⑥ Code-Intel Refresher for a DDD — regenerate code-intel.json from the ⑤-bound repo's source via the BUNDLED engine (scripts/refresh_code_intel.py + ai_ready_helpers.py, pure-stdlib + git, no SwarmAI backend). Narrow refresh mode: NEVER touch the 4 DDD docs. Ships the refresher (capability), never the projection (derived data). TRIGGER: "refresh code intel", "regenerate code graph", "ddd refresher". NOT FOR: generating the 4 DDD docs (that's a full ai-ready-repo run / s_ddd-manager).

Self-maintenance engine: scan 7 dimensions (31 checks), auto-fix mechanical issues, report Found/Fixed/Pending, escalate only judgment calls.

Swarm's media value delivery engine. Transforms messages into optimized media packages (poster, video, narrative, shorts, README) via a 9-stage pipeline (EVALUATE->REFLECT). Message-first, format follows audience. Your message, their attention, the right format. TRIGGER: "pollinate", "make content about", "GTM for", "promote". NOT FOR: autonomous-pipeline, deep-research use cases.

Express THIS DDD's product value to audiences — message-first, format follows — via a bundled, decoupled copy of SwarmAI's real pollinate engine (11 format tracks + scripts + moat gates), sourcing value from the DDD's own ② PRODUCT.md, file-based, no SwarmAI backend. DDD-native decouple of s_pollinate. TRIGGER: "ddd pollinate", "express this ddd's value", "make content from this ddd". NOT FOR: SwarmAI's own s_pollinate (bound to Swarm brand + SwarmWS paths).

Provision and manage DDD projects that conform to the canonical six-section structure — the self-propagation seed. Full lifecycle: CREATE (with context-extraction from a codebase, NOT blank templates), BIND a repo (⑤ delivery contract), LIST / EDIT / RENAME. Pure filesystem, no SwarmAI backend — a DDD carries this so it can create MORE spec-compliant DDDs on Kiro / Claude Code. TRIGGER: "create ddd", "new ddd project", "add a domain", "bind repo to ddd", "provision ddd". NOT FOR: SwarmAI's own project CRUD (native s_project-manager); code (that's s_ddd-pipeline).

Sediment knowledge into THIS DDD's docs — with the FULL routing discipline of SwarmAI's s_persist: a Step-0 admission gate (most facts should NOT be persisted), a governance boundary, content-type routing, and a sanctioned RETIRE/MOVE reverse side. Additive, honors human-authored content, never overwrites judgment. Uses the bundled locked-write engine (scripts/locked_write.py, no SwarmAI backend). TRIGGER: "persist to ddd", "remember in this ddd", "sediment this lesson", "update the ddd docs". NOT FOR: SwarmAI's own MEMORY/EVOLUTION routing (that's native s_persist); behavioral RULES (→ governance, not knowledge).

Backup and restore SwarmAI workspace (memory, knowledge, projects, config, conversations) to/from a private GitHub repository.

Create, evaluate, and improve Agent skills to production quality. TRIGGER: "create a skill", "build a skill", "review skill", "score skill". NOT FOR: skillify-session use cases.

Post-build verification of SwarmAI critical assumptions — streaming, context files, MCP, DailyActivity pipeline.

Deploy, update, manage, and monitor Hive cloud instances. Use when the user wants to create a new Hive, update an existing Hive to a new version, check Hive status, start/stop instances, reset passwords, or manage AWS accounts for Hive deployment.

Bump version across all package files, update CHANGELOG, create git tag, and publish GitHub Release. Ensures no file is missed during version bumps. SUPERSEDED BY: s_swarm-release (which includes build + package + smoke test). Use this skill ONLY for version-bump-without-shipping scenarios. TRIGGER: "bump version only", "version bump", "just tag". NOT FOR: s_swarm-release, s_swarm-build use cases.

Runtime daemon + live-session liveness probe (zero-LLM, deterministic): daemon up, sessions progressing (not wedged — RP41 double-signal), RSS under budget, no unrecovered failure events, deployed commit == expected. Snapshot now; runs as a 15-min job; red → Slack.

Build SwarmAI backend binary (PyInstaller), verify capabilities, deploy to daemon, and restart. Replaces ./prod.sh build with per-stage visibility. TRIGGER: "build", "build backend", "swarm build", "deploy binary". NOT FOR: s_swarm-release, s_swarm-dev use cases.

Check SwarmAI GitHub Actions CI status: list recent runs, diagnose failures, and summarize health. Replaces ad-hoc gh run commands with structured output. TRIGGER: "CI status", "check CI", "is CI green", "CI failures". NOT FOR: pytest use cases.

Manage the SwarmAI launchd daemon: status, stop, start, restart, deploy, logs, and health verification. Replaces manual launchctl and dev.sh daemon commands with structured per-step execution. TRIGGER: "daemon status", "restart daemon", "daemon logs", "stop daemon". NOT FOR: s_swarm-build, s_swarm-hive use cases.

Generate HTML artifact files for human-consumed outputs (reports, reviews, scorecards, comparisons). Dual-consumer protocol: agent self-use = markdown always; human consumption = HTML when content triggers escalation (>100 lines, multi-dimensional, spatial, interactive, shareable). TRIGGER: "generate report", "html report", "comparison table", "scorecard". NOT FOR: files, frontend-design use cases.

Ultra-compressed communication. Cuts tokens ~70% by dropping articles, filler, pleasantries, hedging. All technical substance stays exact. TRIGGER: "caveman", "caveman mode", "be brief", "less tokens". NOT FOR: full use cases.

Structured code review for PRs, files, or diffs with actionable findings. TRIGGER: "review code", "code review", "review PR", "review this file". NOT FOR: web-design-review, simplify use cases.

Governance lifecycle manager for the Three-Layer Governance OS. Six operations: CLASSIFY, CAPTURE, PROMOTE, REFINE, RETIRE, COMPRESS. Manages principles (SOUL), rules (AGENT/STEERING), and gates (hooks/code). Detects capability gaps, captures corrections with bias tags, auto-promotes recurring patterns (3x threshold), and prevents governance bloat. TRIGGER: "add rule", "steeringify". NOT FOR: for one-off file edits (just edit directly).

Create production-grade frontend interfaces with design intelligence database (67 styles, 161 palettes, 57 fonts, 161 industry rules). TRIGGER: "build a landing page", "create a website", "design a UI", "frontend prototype". NOT FOR: web-design-review use cases.

Add, validate, and promote SwarmAI eval golden cases through a 4-gate quality check (schema/duplicate/non-vacuous/privacy). The ONLY sanctioned path to modify the golden set — keeps cases from being 太随便. ADD lands private by default; PROMOTE runs the privacy gate before a case ships public. TRIGGER: "add golden case", "new eval case", "promote case to public", "validate golden set". NOT FOR: running eval (use eval_runner), s_self-evolution governance rules.

Tiered chat experience validator -- quick checks (5min) after every change, full audit (30min) before releases. Covers state machine invariants, SSE pipeline, streaming indicators, queue drain, and regression detection.

Generate DDD Cultivation Weekly Report — summarizes all DDD document changes across all projects for the past 7 days. Shows what was auto-applied, what needs escalation, and per-project DDD health. MBR format. TRIGGER: "DDD weekly", "DDD report", "cultivation report", "DDD 周报". NOT FOR: swarmai-monthly-report, weekly-report use cases.

Score Salesforce opportunities using the MEDDPICC framework (8 dimensions, 100 points). Pulls data from CRM MCP, scores each dimension against evidence, checks stage gates, detects risk signals, and generates prioritized actions with estimated point impact. TRIGGER: "MEDDPICC", "MEDDPICC analysis", "MEDDPICC score", "score this opportunity".

Create, manage, and list scheduled tasks using crontab and launchd (macOS). No external dependencies. TRIGGER: "schedule", "cron", "crontab", "run every". NOT FOR: apple-reminders, outlook-assistant use cases.

Generate SwarmAI Monthly Report — MBR-style health & progress covering all 12 Core Engine subsystems (Memory, Context, Pipeline, DDD, Evolution, Health, Jobs, Code Intel, Skills, Pollinate, Sessions, Git). Outputs to Knowledge/Reports/. TRIGGER: "monthly report", "月报", "SwarmAI health", "system report". NOT FOR: ddd-weekly-report, monthly-report use cases.

Full system health report: desktop overview, worst offenders, SwarmAI resource details, and actionable suggestions. Outputs a structured report in the chat window. TRIGGER: "system health", "mac health", "linux health", "battery check". NOT FOR: cloudwatch-log-analysis, health-check, bsc-security-scanner use cases.

Word document creation, editing, and analysis (.docx files). TRIGGER: any request involving Word documents, tracked changes, or .docx files. DO NOT USE: for plain text or markdown documents.

Translate text, documents, and code between languages. Supports files, inline text, and batch operations. TRIGGER: "translate", "translation", "convert to Spanish", "in Japanese". NOT FOR: summarize use cases.

Create, manage, and query Apple Reminders from the terminal. Syncs to iPhone/iPad/Mac. TRIGGER: "remind me", "add reminder", "create reminder", "my reminders". NOT FOR: for calendar events (use Outlook Assistant), project management, or non-Apple reminder systems.

DOM-based browser automation: navigate websites, read compressed page content, click elements, fill forms, extract data, and take screenshots using Playwright. TRIGGER: "browse", "open website", "browser agent", "web automation". NOT FOR: peekaboo use cases.

Two-phase consulting-grade research reports with professional analysis frameworks. TRIGGER: "consulting report", "market analysis", "competitive analysis", "industry report". NOT FOR: deep-research, narrative-writing use cases.