About recall
recall reconstructs a project's current work status from persisted state, verification and decision ledgers, transcript history, and repository rules. It is for agents or teams resuming work who need an evidence-based account of what was proven, what remains open, and the next step.
Key features
- 01Reads durable project and session state
- 02Checks verification and decision ledgers
- 03Uses transcript history and repository rules
- 04Reports status, proven work, and open threads
- 05Identifies a concrete next action
Use cases
- 01Resume work after a previous session
- 02Report the latest verified result with SHAs
- 03Find promised work that remains unfinished
Install
Copy the command and run it in a terminal.
$
npx skills add https:/ / github.com/ yeachan-heo/ oh-my-claudecode --skill recallAfter installing, type/recallin Claude Code to confirm it loaded.
Details
From the project READMERecall
Reconstruct where the work stands from what was persisted, in order of trust:
- Durable state:
.omc/notepad.md,.omc/project-memory.json, and the per-mode state in both active layouts (ralph/autopilot/team) — the session-scoped.omc/state/sessions/{id}/and the legacy.omc/state/<mode>-state.jsonfallback for runs persisted before a session id existed (docs/HOOKS.md keeps both live) — what was in flight, what iteration, what was blocked. - Ledgers and trails: under OMC's state root — the verification ledger
(
state/verification-ledger.tsv, by default the checkout's.omc/state/;OMC_STATE_DIRor a parent.omc-workspacemoves the root) and everystate/decision-trail-*.tsvfile (what was decided and why); enumerate the glob before reconstructing. - The transcript tail: the last N messages of this session — what was said, what was promised.
- The shared record: the repo's standing rules (AGENTS.md/CLAUDE.md) name project-specific places to check before declaring anything "never recorded".
Report
- Where it stands: one paragraph, from the state files.
- What was last proven: from the ledger, with SHAs.
- Open threads: promised-but-not-done, from the transcript and trails.
- The next action: one concrete step, named — then take it if the user asked for continuation, propose it if they asked for status.
Rules
- Report from the artifacts, not from memory; memory is what the artifacts are for.
- A thread with no artifact behind it is reported as "mentioned but not persisted", never reconstructed from vibes.
- Human rulings encountered in the trails keep their original wording
(
whyskill epistemics).
Comments
Markdown supported