Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create or update continuity ledger for state preservation across clears
.claude/skills/continuity-ledger/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | — | — |
| case-18 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-06 | ✗→✓ | ▲ Improved | — | — |
| case-05 | ✗→✓ | ▲ Improved | — | — |
> Note: This skill is now an alias for /create_handoff. Both output the same YAML format.
Create a YAML handoff document for state preservation across /clear. This is the same as /create_handoff.
First, determine the session name from existing handoffs:
bashls -td thoughts/shared/handoffs/*/ 2>/dev/null | head -1 | xargs basename
This returns the most recently modified handoff folder name (e.g., open-source-release). Use this as the handoff folder name.
If no handoffs exist, use general as the folder name.
Create your file under: thoughts/shared/handoffs/{session-name}/YYYY-MM-DD_HH-MM_description.yaml, where:
{session-name} is from existing handoffs (e.g., open-source-release) or general if none existYYYY-MM-DD is today's dateHH-MM is the current time in 24-hour format (no seconds needed)description is a brief kebab-case descriptionExamples:
thoughts/shared/handoffs/open-source-release/2026-01-08_16-30_memory-system-fix.yamlthoughts/shared/handoffs/general/2026-01-08_16-30_bug-investigation.yamlCRITICAL: Use EXACTLY this YAML format. Do NOT deviate or use alternative field names.
The goal: and now: fields are shown in the statusline - they MUST be named exactly this.
yaml--- session: {session-name from ledger} date: YYYY-MM-DD status: complete|partial|blocked outcome: SUCCEEDED|PARTIAL_PLUS|PARTIAL_MINUS|FAILED --- goal: {What this session accomplished - shown in statusline} now: {What next session should do first - shown in statusline} test: {Command to verify this work, e.g., pytest tests/test_foo.py} done_this_session: - task: {First completed task} files: [{file1.py}, {file2.py}] - task: {Second completed task} files: [{file3.py}] blockers: [{any blocking issues}] questions: [{unresolved questions for next session}] decisions: - {decision_name}: {rationale} findings: - {key_finding}: {details} worked: [{approaches that worked}] failed: [{approaches that failed and why}] next: - {First next step} - {Second next step} files: created: [{new files}] modified: [{changed files}]
Field guide:
goal: + now: - REQUIRED, shown in statuslinedone_this_session: - What was accomplished with file referencesdecisions: - Important choices and rationalefindings: - Key learningsworked: / failed: - What to repeat vs avoidnext: - Action items for next sessionDO NOT use alternative field names like session_goal, objective, focus, current, etc. The statusline parser looks for EXACTLY goal: and now: - nothing else works.
IMPORTANT: Before responding to the user, you MUST ask about the session outcome.
Use the AskUserQuestion tool with these exact options:
Question: "How did this session go?"
Options:
- SUCCEEDED: Task completed successfully
- PARTIAL_PLUS: Mostly done, minor issues remain
- PARTIAL_MINUS: Some progress, major issues remain
- FAILED: Task abandoned or blockedAfter the user responds, mark the outcome:
bash# Mark the most recent handoff (works with PostgreSQL or SQLite) PROJECT_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || echo "${CLAUDE_PROJECT_DIR:-.}") cd "$PROJECT_ROOT/opc" && uv run python scripts/core/artifact_mark.py --latest --outcome <USER_CHOICE>
After marking the outcome, respond to the user:
Handoff created! Outcome marked as [OUTCOME].
Resume in a new session with:
/resume_handoff path/to/handoff.yaml/clearEach compaction is lossy compression—after several compactions, you're working with degraded context. Clearing + loading the handoff gives you fresh context with full signal.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | pass→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 17 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +64 percentage points is the difference between those two pass rates over the 17 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.