| name | digest |
| description | Capture meaningful session state (current cursor, in-flight work, live hypotheses, open questions, do-NOT-carry-forward, success criteria) to state/digests/latest.md so the next session — including one after auto-compaction — can re-enter without re-discovering context cold. Distinct from /learnings (which captures surprises). Validates against docs/contracts/digest-schema.json. |
| allowed-tools | Bash, Read, Write, Edit |
| last_verified | "2026-05-14T00:00:00.000Z" |
| karpathy_rules | ["R1-think-before-coding","R4-goal-driven-execution"] |
/digest
Captures the meaningful state of the current session to
state/digests/latest.md. The .claude/hooks/sessionstart-digest.sh hook
reads it on the next session start and emits it as additionalContext for
the model. Pairs with .claude/hooks/precompact-digest.sh which provides
a metadata-only fallback when /digest isn't invoked before auto-compaction.
When to run
- Before compaction is imminent (context feels >60% full).
- At a natural breakpoint — after finishing a feature, before a risky
operation, before handing off to another agent.
- Before launching long-running background work.
Do NOT invoke on every message or tool call.
What it captures
---
schema_version: 1
session_id: <session_id>
written_at: <RFC3339 UTC>
trigger: digest-skill
branch: <git branch>
head_sha: <short SHA>
head_subject: <commit subject>
open_pr: <"#N" or null>
last_model_update: <RFC3339 UTC>
success_criteria:
- criterion: "<testable assertion for the Next action>"
status: pending | in-progress | achieved | abandoned
evidence: "<file:line | PR# | metric path | null>"
---
ONE imperative sentence. Maps 1:1 to `success_criteria` entries.
Work currently mid-execution. file/feature, step N of total, next sub-step.
Working hypotheses to inherit. *Unconfirmed*; don't promote to MEMORY.md yet.
Numbered. Questions you would ask the user if they walked in now.
**Highest-leverage field.** Things the next session must NOT re-propose.
When a prior digest exists, this section reports the status delta:
- ✅ achieved: <prior criterion> — evidence: <file:line | commit | PR>
- ⏳ in-progress: <prior criterion> — last touched: <where>
- ⛔ abandoned: <prior criterion> — reason: <one sentence>
How to run
- Read existing
state/digests/latest.md if present — preserve current sections.
- Karpathy R4 — review/mark prior success criteria.
- Capture git state via Bash (
git rev-parse, git log -1, gh pr view).
- Synthesize content; derive 1–3 testable
success_criteria from Next action.
- Write to
state/digests/latest.md. Set trigger: digest-skill + last_model_update.
rm -f state/digests/.activity-counter to reset the staleness nudge.- Validate:
bash .claude/hooks/digest-validate.sh. Non-zero = fix.
- Report:
Digest updated: <branch>@<sha> · next: <one-line> · criteria: <N>.
Driving criteria autonomously with /goal
After writing the digest, consider /goal <condition> (native Claude
Code v2.1.139+) for substantial autonomous work. /goal mechanizes
Karpathy R4: a small fast model evaluates after every turn whether the
condition holds and either fires another turn or clears the goal.
Use /goal when the Next action has:
- A verifiable end state (build green, tests pass, file count under budget)
- 5+ minutes of expected autonomous work
- An evaluator-checkable result (checked against the transcript — no
tool calls)
Skip /goal for short tasks, visual/UX judgement, or ambiguous specs.
When /goal lands a "yes," the next /digest should mark the
corresponding success_criteria entry as achieved with the /goal
evidence.
Anti-patterns
- Transcript capture — git log is the transcript.
- Empty digest — if nothing changed, prior digest still applies.
- Forgetting "Do NOT carry forward".
Related
/learnings — captures surprises into docs/solutions/./correct — captures user corrections as permanent CLAUDE.md rules..claude/hooks/{precompact,sessionstart}-digest.sh — auto-fallback + read-back.state/digests/README.md — directory layout.
