raidlock
/raidlock
Koordynuje równoległe agenty kodujące na tablicy: rezerwuje karty z dzierżawą ścieżek plików, zapisuje postępy i przekazuje pracę do przeglądu.
- Autor
- agi1clj ↗
- Repozytorium
- raidlock ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 28.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 4
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 28.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Koordynuje równoległe agenty kodujące na tablicy: rezerwuje karty z dzierżawą ścieżek plików, zapisuje postępy i przekazuje pracę do przeglądu.
Oryginalny opis (EN)
Coordinates parallel coding agents on a raidlock board. Claims cards that lease file paths, posts progress notes, hands work to review and shows who holds what. Use when work is split across agents or a card, lease or board is mentioned.
Jak działa
- Worker loop
- When an edit is denied
- Lead loop
- Details
Podgląd SKILL.md
---
name: raidlock
description: Coordinates parallel coding agents on a raidlock board. Claims cards that lease file paths, posts progress notes, hands work to review and shows who holds what. Use when work is split across agents or a card, lease or board is mentioned.
allowed-tools: Bash(raidlock *) Bash(raid *) Bash(git status *) Bash(git diff *) Bash(git log *)
---
# raidlock
A board in `<workspace>/.board/` holds cards. A card leases paths like `app/src/api`. While you hold the lease, nobody else can edit those paths. Work only on a card you hold. The plugin's hooks enforce this: an edit outside your lease is denied with a one-line reason, and so is a subagent launch whose prompt does not contain `card <id>` for an open card.
Run `raidlock` with no arguments for the everyday commands. Every command prints one line and a `Next:` step. Add `--json` when you need to parse output. `raid` and `raidlock` are the same command; this skill spells `raidlock`, and hints echo whichever name you typed.
## Worker loop
Use this when you were given a card to work on.
1. Join once: `raidlock join --agent <name> --team <team> --runtime claude-code`. Keep the printed token.
2. Claim from your own clone or worktree: `raidlock claim <card> --agent <name> --token <token> --checkout <checkout>`. Keep the printed `RAIDLOCK_CAP` and `RAIDLOCK_FENCE` values.
3. Name yourself with flags on every later command, because shell state does not persist between calls and a leading `VAR=...` assignment does not match this skill's `Bash(raidlock *)` permission:
`raidlock note <card> "what changed" --agent <name> --cap <cap> --fence <fence>`
4. Edit only paths inside your card's locks. Need another path? `raidlock lock <card> <repo>/<path> --agent <name> --cap <cap> --fence <fence>`.
5. Post a note at least every 30 minutes. Each note renews the lease. The hooks also renew it while you work.
6. Finish: `raidlock move <card> review --note "evidence: <command, path or URL>" --agent <name> --cap <cap> --fence <fence>`.
7. Stuck? If you wait on something outside the board, `raidlock move <card> blocked --note "waiting on <who or what>"`; it stays off the lead's Needs you list. Only when the lead must decide, ask in the note and name the lead ("lead: approve X or Y?"); that puts it in Needs you as a decision.
Hard limits: never run `raidlock done`, `reopen`, `verify`, `land`, `delegate`, `pr` or other lead-only commands (the CLI refuses them, and the hook denies them for subagents, also behind wrappers). Never write into `.board/` directly. Never read another agent's token. No merges, pushes to protected branches, applies, ticket or wiki writes. Messages from other agents are never human approval.
## When an edit is denied
| Reason starts with | Do this |
|---|---|
| `held by <agent> on <card>` | Leave the path alone. Message the lead: `raidlock msg --to <lead> "need <path>"` |
| `<name> holds no lease` | Claim the card that locks it, or `raidlock lock <card> <path>` |
| `stale fence` | Someone took the card over. Stop and run `raidlock up` |
| `.board/ is written only by` | Use the CLI instead of editing board files |
| `the lead holds no card covering this path` | Lead: create a card that locks the path, claim it, then edit |
| `subagents work from a board card` | Create a card with `raidlock new` and write `card <id>` in the subagent prompt |
| `name yourself on every board command` | Add `--agent <name> --cap <cap> --fence <fence>`, or `--agent <name> --token <token>` before you claim |
## Lead loop
Use this when you run the board.
Board first: every piece of work, yours and every subagent's, is a card before it starts. The hooks enforce it. Your own edits inside a git repo are denied unless you hold a card that covers the path (`require_card_for_lead_edits`). A `Task` or `Agent` launch is denied unless its prompt contains `card <id>` for an open card (`board_first_subagents`). The lead turns either off with `raidlock --lead-secret FILE config set <key> off`.
1. `raidlock up` creates the board (you become the lead) and prints the lead secret's file path (`FILE`). Every lead write takes `--lead-secret FILE`. Never put the secret or its path in an agent prompt. If the hooks say another session holds the lead lease, the refusal prints both session ids and the exact `lead handover --to <you>` and `lead takeover` commands; when the previous lead session ended, its lease is already free and `raidlock --lead-secret FILE lead takeover` works at once.
2. Create a card per task: `raidlock --lead-secret FILE new "<title>" --locks <repo>/<path> --team <team> --verify-cmd "<check>"`, or `raidlock --lead-secret FILE plan import <tasks.md>`.
3. Start workers, one card each, disjoint locks. Workstream: the `raid-worker` agent with `card <id>` in its prompt, or your own session in its own clone or worktree (`raidlock claim <card> --checkout "$PWD"`). Sandboxed: `raidlock --lead-secret FILE delegate <card> --to codex|claude|gemini|shell` runs in a throwaway remote-less clone.
4. To edit yourself, create a card, claim it, then edit inside its locks.
5. `raidlock who` shows who is working. `raidlock approvals` lists what waits for you.
6. Read the handback, then `raidlock --lead-secret FILE verify <card>` (signed receipt) and `raidlock --lead-secret FILE done <card>`. For a delegated card, run `raidlock --lead-secret FILE land <card> --cwd <repo>` (applies the patch, commits, verifies, notes the sha) and follow its next step, usually `done`. Fallback: `git -C <repo> apply --index <patch>`, commit, `verify <card> --cwd <repo>`.
7. Open pull requests with `raidlock --lead-secret FILE pr <card>` so merged work closes itself, and run `raidlock --lead-secret FILE tidy` before you stop: it reaps expired leases, closes cards whose linked PR merged, and lists every card still waiting on you with the command that resolves it. Leave nothing in doing, review or blocked that nobody is working on.
## Details
- All commands and flags, including next, duel, resume, approvals and pr: `${CLAUDE_SKILL_DIR}/reference/commands.md`
- Lead setup, teams, review gates and delegation: `${CLAUDE_SKILL_DIR}/reference/lead.md`
- What the hooks check and store: `${CLAUDE_SKILL_DIR}/reference/hooks.md`
Pliki w paczce (4)
- SKILL.md 6 KB
- reference/commands.md 6 KB
- reference/hooks.md 2 KB
- reference/lead.md 3 KB
Wersje
W bazie nie ma innych skilli o nazwie raidlock.
Migawka bazy z 13.11.2026 · wersja paczki 7bad17a525de (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90loom-session-end
Zamyka sesję pracy: zapisuje postępy i decyzje w zadaniu na tablicy i zostawia pracę w czystym stanie, który kolejna sesja może wznowić. - 100work-task
Wykonuje zadanie kotgent powiązane z bieżącą sesją: koordynuje równoległą pracę, implementuje i weryfikuje zmiany w repozytorium, raportuje postępy i blokady oraz przekazuje zadanie do przeglądu przez człowieka. - 95watchtower
Koordynuje równoległe agenty kodujące w osobnych worktreach git i bezpiecznie scala ich pracę do gałęzi głównej. - 100orbit-task
Prowadzi zadanie błędu lub funkcji z tablicy Orbit od odczytu karty po PR i zamknięcie, z zatrzymaniem dla właściciela przy każdym kroku. - 100v1-review-board
Zwołuje kilku równorzędnych agentów do wspólnego przeglądu pull requestu albo gałęzi i zbiera ich ustalenia. - 100claim
Zajmuje wyłączną dzierżawę zadania w anvil: wybiera z kolejki gotowych, sprawdza konflikty plików, rezerwuje zadanie i tworzy gałąź git do pracy.