Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Development skill from everything-claude-code
.claude/skills/cc-skill-strategic-compact/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | — | — |
| case-03 | ✗→✓ | ▲ Improved | — | — |
| case-02 | ✗→✗ | = Same ✗ | — | — |
| case-09 | ✗→✗ | = Same ✗ | — | — |
| case-07 | ✗→✗ | = Same ✗ | — | — |
Condense context after a verified phase, with enough durable information to resume the same task. A tool-call count is only a reminder; it cannot measure context quality or decide when compaction is safe.
Use when exploration has finished, a tested milestone has landed, or a long conversation contains repeated or obsolete reasoning. Do not compact in the middle of an unresolved tool call, an unrecorded approval decision, or a failure whose exact evidence has not been captured.
/compact exists in every client or manipulate internal context files.textObjective and done condition: User constraints and authorization boundaries: Source revision, branch and local changes: Completed outcomes with verification: Outstanding work and known failures: Running process/tool identifiers: Artifacts and exact paths needed next: Next action, expected result and fallback:
For a search feature whose source PR merged while canonical synchronization is pending, preserve the source merge SHA, canonical PR number and expected generated files. Record tests as passed only for the revision actually tested. The next action is to verify the canonical PR and updated main; it is not to merge the source PR again. If a browser check still needs an unlocked desktop, keep that condition visible while continuing independent code review.
Expected outcome: the next context resumes outstanding verification, preserves unrelated local edits and does not claim publication from a source merge.
bashbash skills/cc-skill-strategic-compact/suggest-compact.sh 50
The helper accepts an observed tool-call count and optional threshold (default 50), and prints a reminder at the threshold and every 25 calls after it. COMPACT_TOOL_COUNT and COMPACT_THRESHOLD can be supplied by an explicit caller adapter. It stores no counter, estimates no token usage and never performs compaction. With no count it exits quietly.
Earlier versions incremented a shared user-state counter across unrelated sessions. That counter is no longer read or written; existing custom wrappers must pass a session count explicitly. Existing state files are left untouched. This skill does not install or edit a hook.
| 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-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
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. The headline lift of -20 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.