Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manages Claude Code sessions with naming, checkpointing, and resume strategies. Use when organizing long-running work or resuming across sessions.
.claude/skills/athola-session-management/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 71% | 0% |
Claude Code supports named sessions for better workflow organization. Use this skill to manage complex, long-running work across multiple sessions.
| Command | Description | |---------|-------------| | /rename | Name the current session (auto-generates name if no argument given, 2.1.41+) | | /resume | Resume a previous session (REPL) | | claude --resume <name> | Resume from terminal |
Name debug sessions for easy resumption:
# Start debugging
/rename debugging-auth-issue
# ... work on the issue ...
# If you need to pause, session is auto-saved
# Resume later:
claude --resume debugging-auth-issueCreate checkpoints during long feature work:
# After completing milestone 1
/rename feature-x-milestone-1
# Continue in new session
# Reference old session if neededFor complex PR reviews that span multiple sittings:
# Start review
/rename pr-review-123
# Take breaks without losing context
# Resume:
claude --resume pr-review-123Sessions are automatically linked to PRs when created via gh pr create. Resume PR-specific sessions later:
# Resume session for a specific PR
claude --from-pr 156
claude --from-pr https://github.com/org/repo/pull/156
# Workflow: review → pause → resume with full context
/rename pr-review-156
# ... review work ...
# Later:
claude --from-pr 156When investigating issues that may require research:
# Start investigation
/rename investigate-memory-leak
# Pause to gather more info externally
# Resume with full context:
claude --resume investigate-memory-leakThe /resume screen provides:
/keybindings):Claude Code now shows a resume hint when you exit, displaying the command to continue your conversation. This makes session resumption more discoverable: users no longer need to know about --resume beforehand.
Use descriptive, hyphenated names:
| Pattern | Example | Use Case | |---------|---------|----------| | debugging-<issue> | debugging-auth-401 | Bug investigation | | feature-<name>-<milestone> | feature-search-v2 | Feature development | | pr-review-<number> | pr-review-156 | PR reviews | | investigate-<topic> | investigate-perf | Research | | refactor-<area> | refactor-api-layer | Refactoring work |
Name sessions when:
Unnamed sessions are eventually garbage collected. Named sessions persist longer. Periodically clean up old named sessions you no longer need.
Combine session management with other Sanctum skills:
Skill(sanctum:git-workspace-review) to capture context/rename <descriptive-name>claude --resume <name>If a named session isn't appearing in /resume:
/resume screen to browse available sessionsIf you see duplicate session entries when resuming in VS Code:
If context seems incomplete or resume is slow:
--resume via stat-based session loading with progressive enrichment, especially impactful for users with many sessions. Also fixes hangs when resuming sessions with corrupted transcript files (parentUuid cycles).once: true hooks: saved_hook_context loading is now optimized/catchup to refresh git state/debug (Claude Code 2.1.30+) for session troubleshooting diagnosticsSkill(sanctum:git-workspace-review) if neededPreviously, disconnecting from a terminal on macOS could leave orphaned Claude Code processes running. This is now fixed in 2.1.46+. If you encounter stale CC processes on older versions, manually check and kill them:
bash# Find orphaned claude processes ps aux | grep -i claude | grep -v grep
Claude now automatically records and recalls memories as it works. Session summaries, key results, and work logs are captured implicitly and recalled in future sessions. This provides passive cross-session continuity without manual checkpoint management.
--resume now re-uses the --agent value from the previous conversation by default. Agent-specific workflows that are resumed will continue with the same agent configuration without needing to re-specify it.
# Start with a specific agent
claude --agent my-agent
# Resume later — my-agent is automatically used
claude --resume/catchup - Refresh context from git changes/clear - Start fresh sessionSkill(sanctum:git-workspace-review) - Capture repo context/rename using a hyphenated descriptive namematching one of the documented patterns (e.g. debugging-*, feature-*, pr-review-*)
claude --resume <name> without loss ofprior context (verified by presence of earlier conversation)
with troubleshooting steps (check typos, session expiry)
<number> correctly reopens the session associated with that PR
git-workspace-review) is re-run and its output confirms current repo state
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 6,668 | 3,999 | -40% | 1 | 1 | 0% | 1,180 | 2,479 | +110% | 0 | 0 | — |
case-02 | fail→pass | 7,325 | 4,740 | -35% | 1 | 1 | 0% | 1,153 | 2,538 | +120% | 0 | 0 | — |
case-03 | fail→pass | 17,481 | 9,129 | -48% | 1 | 1 | 0% | 2,737 | 3,335 | +22% | 0 | 0 | — |
case-04 | pass→pass | 3,619 | 2,564 | -29% | 1 | 1 | 0% | 640 | 2,235 | +249% | 0 | 0 | — |
case-05 | pass→pass | 5,424 | 1,890 | -65% | 1 | 1 | 0% | 744 | 2,087 | +181% | 0 | 0 | — |
case-06 | pass→pass | 3,371 | 2,691 | -20% | 1 | 1 | 0% | 506 | 2,229 | +341% | 0 | 0 | — |
case-07 | fail→pass | 8,656 | 3,441 | -60% | 1 | 1 | 0% | 1,411 | 2,331 | +65% | 0 | 0 | — |
case-08 | fail→pass | 10,534 | 5,139 | -51% | 1 | 1 | 0% | 1,717 | 2,542 | +48% | 0 | 0 | — |
case-09 | fail→pass | 9,228 | 3,693 | -60% | 1 | 1 | 0% | 1,422 | 2,430 | +71% | 0 | 0 | — |
case-10 | fail→pass | 15,787 | 6,035 | -62% | 1 | 1 | 0% | 2,478 | 2,756 | +11% | 0 | 0 | — |
case-11 | fail→pass | 20,979 | 3,423 | -84% | 1 | 1 | 0% | 3,426 | 2,347 | -31% | 0 | 0 | — |
case-21 | fail→pass | 15,259 | 4,158 | -73% | 1 | 1 | 0% | 2,452 | 2,515 | +3% | 0 | 0 | — |
case-12 | fail→pass | 5,832 | 2,761 | -53% | 1 | 1 | 0% | 858 | 2,186 | +155% | 0 | 0 | — |
case-13 | pass→pass | 15,194 | 4,663 | -69% | 1 | 1 | 0% | 2,415 | 2,615 | +8% | 0 | 0 | — |
case-14 | pass→pass | 10,546 | 3,160 | -70% | 1 | 1 | 0% | 1,842 | 2,160 | +17% | 0 | 0 | — |
case-15 | fail→pass | 21,080 | 2,096 | -90% | 1 | 1 | 0% | 3,538 | 2,084 | -41% | 0 | 0 | — |
case-16 | fail→pass | 20,592 | 2,168 | -89% | 1 | 1 | 0% | 3,423 | 2,078 | -39% | 0 | 0 | — |
case-17 | fail→pass | 15,333 | 2,516 | -84% | 1 | 1 | 0% | 2,621 | 2,112 | -19% | 0 | 0 | — |
case-18 | pass→pass | 7,227 | 3,817 | -47% | 1 | 1 | 0% | 1,092 | 2,499 | +129% | 0 | 0 | — |
case-19 | pass→pass | 13,629 | 7,163 | -47% | 1 | 1 | 0% | 1,942 | 2,922 | +50% | 0 | 0 | — |
case-20 | pass→pass | 12,761 | 1,765 | -86% | 1 | 1 | 0% | 1,966 | 2,082 | +6% | 0 | 0 | — |
case-22 | fail→pass | 9,341 | 3,581 | -62% | 1 | 1 | 0% | 1,412 | 2,355 | +67% | 0 | 0 | — |
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 +59 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.