From springfield
Use Springfield recover to diagnose a stuck batch or failed slice and restore a safe next step.
How this skill is triggered — by the user, by Claude, or both
Slash command
/springfield:recoverThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use Springfield recover to diagnose a stuck batch or failed slice and restore a safe next step.
Use Springfield recover to diagnose a stuck batch or failed slice and restore a safe next step.
Source: builtin/springfield.md
Built-in Springfield playbook.
Run springfield version first. It prints one of:
springfield vX.Y.Z — a released build.springfield dev — a local source build (e.g. go install .).Then:
brew install brentguistwite/tap/springfieldspringfield_<version>_linux_<arch>.tar.gz from the GitHub Releases page and put the springfield binary on PATH.go install . inside the Springfield repo (no Windows release tarballs are published yet).springfield dev, this is a local development build. Continue without a floor check — the user is responsible for keeping it current.v from the reported version), tell the user to upgrade, then stop:
brew upgrade springfieldspringfield_<version>_linux_<arch>.tar.gz from the GitHub Releases page and replace the binary on PATH.go install . inside the Springfield repo (no Windows release tarballs yet).Do not try to work around a missing or too-old CLI — surface the exact command above instead. (A plugin older than the CLI is fine and needs no action; the CLI stays backward-compatible with older skills within a major version.)
Reads are allowed — recover and status flows specifically inspect .springfield/run.json and per-plan prd.json. Never write, edit, or delete files under .springfield/. That directory is Springfield's state — the CLI is your only interface for mutating it. Writing there directly will abort the current batch. This applies regardless of which agent is invoking the skill.
Recover a Springfield batch that is stalled, blocked, or has a failed slice.
Read project guidance from AGENTS.md first, then CLAUDE.md, then GEMINI.md when present.
Run springfield status to see the active batch, current phase, and slice statuses.
For a failing or stalled plan, run springfield recover --diagnose --plan <plan-id> first. The CLI surfaces dirty-worktree state, exit_reason, orphan-plan handling, and the available recovery actions — interpret that output rather than re-deriving it from raw files.
Also read .springfield/run.json for the last checkpoint and last known error.
Identify which plan and which story failed or stalled. Check:
run.jsonpasses state in each plan's prd.json (.springfield/plans/<plan-id>/prd.json)Propose the safest concrete next step:
springfield start to resume from cursor. Already-passed stories (passes: true) are skipped on re-entry — prd.json is preserved across retries.springfield plan --replace to start fresh with a new batch.Story-level retry is not supported in v1; recovery operates at plan grain (retry plan or retry merge).
Prefer recovery and continuation over starting a fresh plan unless the existing state cannot be salvaged. Keep Springfield as the only user-facing surface.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub brentguistwite/springfield --plugin springfield