Apocrathia/homelab

a2a

Invoke homelab kagent agents through the LiteLLM A2A broker (list agents, send messages, stream, poll tasks). Use when delegating to git-agent, homelab-agent, hermes-agent, or other registered A2A peers — not via MCP.

原文の言語: 英語

reconcile-context

Reconcile AGENTS.md and .agents/context/ against the repo: fix broken links, catalog drift, and routing mismatches. Use when asked to sync, reconcile, or update the agent context.

原文の言語: 英語

reconcile-docs

Close the docs/backlog side of a lap: fix behavior docs the change made wrong; delete satisfied issues/plans; fix backlinks. Use after implement, before draft-commit — not for AGENTS/.agents context drift.

原文の言語: 英語

self-improve

Well-lit path from blank start to merged contribution: discover ranked work, execute one launch brief from find-work, reconcile docs and context, ship, cleanup. Use when the operator says self-improve, /self-improve, run the work graph or work loop, go…

原文の言語: 英語

alignment

Interview the operator until shared understanding. Gated branches one at a time; independent discrete choices may share one structured-question turn. Use when stress-testing an idea, design, or requirement; resolving open questions before acting; or when the…

原文の言語: 英語

autoresearch

Idle-only tier-8 research lap: run bounded, metric-driven experiments from an operator-approved contract; ship writeups under docs/research/ only. Experimental code does not enter GitOps. Use when the operator says autoresearch, or find-work surfaces a tier-8…

原文の言語: 英語

cleanup-worktrees

Inventory and remove unused git worktrees and local branches that are fully merged (or content-equivalent) on main. Required before creating a new worktree (see worktrees.md). Also use when the user asks to clean up worktrees, prune merged branches, remove…

原文の言語: 英語

clock-out

Tear down the agent worktree and stop after a merged MR. Use when the user says clock out, /clock-out, clean up your workspace, or sends the post-merge dismissal.

原文の言語: 英語

create-agent

Create a new agent persona under .agents/agents/. Use when the user says create agent, new agent, add a persona, or when a recurring delegation pattern needs its own subagent charter.

原文の言語: 英語

create-skill

Create a new skill under .agents/skills/. Use when the user says create skill, new skill, add a skill, or when a repeated multi-step procedure should become a reusable, invocable workflow instead of being re-explained each session.

原文の言語: 英語

implement-change

Orchestrate one Launch-brief lap: plan if needed, implement↔reviewer pair, verifier arbiter, domain review when warranted. Use after find-work selects a brief.

原文の言語: 英語

integrate-upstream

Pull changes from the prime-context core into this project's existing .agents/ tree — diff shared rules, skills, and scripts, reconcile layer 1 and 2 changes, and preserve project-specific content. Use when the user says sync upstream, integrate upstream…

原文の言語: 英語

retrospective

Structured post-work retrospective: review outcomes, mine git history for context-relevant changes outside the session, extract lessons, classify them, and route improvements — local context edits, rule/skill/enforcement proposals, or upstream contributions…

原文の言語: 英語

run-loop

Constant-loop orchestrator: walk find-work Launch briefs in scout, unattended, or attended mode. Never a persona. Use for Cron scouts and AFK/attended lap walks — not for inventing work.

原文の言語: 英語

ship-work

Commit, push, open a GitLab MR, then watch it until merged or dismissed. Use after review-loop is clean, when asked to ship, open an MR, or watch a branch MR.

原文の言語: 英語

find-work

Rank read-only work from the local backlog and live signals; emit Launch briefs. Use when finding, prioritizing, or picking the next agent lap.

原文の言語: 英語

mcp-deployment

Deploy MCP servers via kmcp (kagent MCPServer) and integrate with LiteLLM in the homelab. Covers MCPServer CRD, transport selection, secret mounts, and litellm.yml / RemoteMCPServer wiring. Use when adding MCP servers under…

原文の言語: 英語

architecture-review

Explore the codebase for architectural friction (shallow modules, coupling, test gaps) and file local markdown issues under docs/issues/. Use when improving codebase shape, before large refactors, or when find-work should surface design-debt issues.

原文の言語: 英語

draft-commit

Handoff a green-enough lap: confirm review evidence, analyze git status/diff, draft Conventional Commit + optional draft MR body, list include/exclude files. Draft-only by default; commits/pushes only when the operator authorizes this lap.

原文の言語: 英語

file-issue

Create or update local issues under docs/issues/; use when filing a gap out of scope for the current change.

原文の言語: 英語

review-loop

Local verify loop before draft-commit: format, lint, link/discovery checks, helm/kustomize renders, Trivy on changed paths. Cap ~5 iterations. Use after implement and reconcile, before proposing a commit.

原文の言語: 英語

prototype

Build a throwaway prototype to answer a design question. Use when the user wants to sanity-check whether a state model or logic feels right, explore what a UI should look like, or says prototype or /prototype.

原文の言語: 英語

helm-deployment

Deploy Helm charts for the homelab GitOps environment using Flux, Kustomize, and generic-app. Covers chart review, HelmRelease configuration, Authentik integration, networking, and post-deploy validation. Use when deploying a new app via Helm, updating…

原文の言語: 英語

watch-mr

Babysit one open GitLab MR: unresolved threads, failing CI, conflicts, approvals, draft status. Prefer GitLab MCP. Never merge or approve without explicit ask; commit/push/undraft only when ship-authorized. Use for tier-3 MR maintenance laps.

原文の言語: 英語

cnpg-logical-database-restore

Perform logical backup and restore of CloudNativePG PostgreSQL databases using cnpg-data-extract and cnpg-data-restore charts. Covers Flux suspend/resume, pg_dump/pg_restore workflow, and Longhorn PVC handling. Use when restoring or migrating CNPG databases…

原文の言語: 英語

generic-app-longhorn-restore

Restore Longhorn-backed volumes for generic-app Helm workloads. Covers Retain policy, fromBackup, helm template PV/PVC re-render, and scale-down/up ordering. Use when recovering faulted or missing Longhorn volumes for generic-app deployments with…

原文の言語: 英語