Skip to main content
On this page

Reference

Generated from the repository source. To refresh after editing any source file:

Shell
npm run docs:reference

Browse by category

Category Count What
Skills 8 Polymorphic skills users invoke directly
Agents 9 Subagents called by skills (never invoked directly)
MCPs 9 MCP configs (env-var driven; zero tokens stored)
Scripts 11 Programmatic helpers (Python + JS)
Hooks 1 PreToolUse / PostToolUse / SessionStart
Shared 6 top-level + 4 subgroups Constitution, advisor, question-first, decision-log schema, edit-format, plan-act-mode, personas, workflows, input-classifiers, guidelines
Agent envs 4 Claude / Cursor / Codex / Junie wrappers

Skills (entry points)

Skill Triggers on Primary sub-flows
adk-implement implement / build / add / write / ship from-jira, from-issue, from-tdd, from-confluence, from-slack-thread, greenfield
adk-review review / audit / look-at / sanity-check review-pr, review-code-changes, review-doc, review-comments, audit-repo, audit-pr
adk-investigate investigate / why / debug / RCA / root-cause incident, rca, experiment, datadog, mixpanel, statsig, snowflake, looker
adk-document document / write / draft / summarize runbook, ADR, RCA, PR body, commit msg, changelog, diagram, readme, migration guide, api reference, experiment report, incident summary
adk-sync publish / sync / push-to / fetch read-confluence/jira/gdoc/gh-pr/gh-issue/slack, write-* (same targets)
adk-setup set-up / configure / refresh-metadata —init, —enrich, —check, —diff
adk-improve improve / learn / refresh-metadata defaults / metadata / both
adk-explain I-don’t-know / explain / help-me-decide advisor hand-off

Subagents (called by skills)

Subagent Persona
adk-agent-code-reviewer Findings-first, severity-tiered, evidence-quoted
adk-agent-security-reviewer Adversarial, threat-modeled, boundary-aware
adk-agent-investigator Two-source minimum, confidence-stated, lowest-blast-radius
adk-agent-doc-writer Reader-first, evidence-cited, no filler
adk-agent-doc-reviewer Stale vs wrong vs incomplete; never rewrites voice
adk-agent-implementer Smallest correct change; SEARCH/REPLACE block discipline
adk-agent-test-engineer Behavior-named tests; fail-first then green
adk-agent-context-gatherer Parallel link follower; one hop only; haiku for cost
adk-agent-explainer Teaches, doesn’t pick for the user

MCPs

MCP Status Auth
adk-mcp-github required for code skills PAT (fine-grained → classic fallback) or OAuth
adk-mcp-datadog required for investigate API + APP key
adk-mcp-statsig required for investigate Console API key
adk-mcp-atlassian required for docs/Jira API token via uvx mcp-atlassian
adk-mcp-mixpanel optional OAuth on first connect
adk-mcp-slack optional Sources $SLACK_CREDENTIALS_FILE
adk-mcp-snowflake optional Snowflake user/password or SSO
adk-mcp-looker optional API3 client id/secret
adk-mcp-rag optional RAG_MCP_URL + bearer token

How to refresh

Edit the source file (skill / agent / MCP / script / shared / hook) and re-run:

Shell
npm run docs:reference

This wipes and regenerates everything under docs/reference/{skills,agents,mcp,scripts,hooks,shared,agent-envs}/ from the v3 source.