Report per-agent token usage for the current Claude Code session (default), all sessions (--all), or a time window (--since {h|d|w}). Reads .claude/.agent-tokens.jsonl produced by the SubagentStop hook and prints a fixed-width table with dollar estimates per Anthropic public pricing.
EnglishAuthor a new plan-doc at docs/plans/active/ .md from a one-line request — copy the template, set the slug to the basename, fill every section (N/A — , never delete), include the Perf-review-system-gates section when the diff would touch the project's perf-gated core source tree, keep the three mandatory Verification/Out-of-scope items the template bakes in (doc-validation suite, grill-with-docs stress-test, deferral residue-sweep), then git add + wip(plan) commit immediately (working-tree-only plans are lost on checkout). Use when the user says "write a plan", "draft a plan", "new plan", "plan for X", or "plan this". The authoring counterpart to grill-with-docs (the stress-test). Read-write — creates + commits the plan file.
EnglishThe deterministic report-shape mechanics of a CodeRabbit-triage round — the triage-table example, the per-finding N handoff-packet skeleton (VALID + REJECTED worked examples), and the reply-to-bot lines. Invoked by coderabbit-triage (which keeps the judgment: the 19-rule override table, path→agent routing, severity classification, watcher-invocation, hand-back contract). Use when formatting a CodeRabbit triage report or a per-finding handoff packet.
EnglishThe deterministic mechanics of a Smatchet debug investigation — the NDJSON [temp-debug] instrumentation recipe, build + exe-staleness commands, the unified-CLI run reference, log/evidence reading (jq), sanitizer setup, cleanup, and the two report-shape templates. Invoked by debug-detective (which keeps the judgment: hypotheses, metric choice, the wait-for-feedback loop, hand-off). Use when instrumenting, reading a debug log, cleaning up [temp-debug] markers, or formatting a debug report.
EnglishDrain the harness auto-memory inbox into the repo. Triages every memory file under ~/.claude/projects/ /memory/ (except MEMORY.md) into implement / backlog / toss per docs/agent-rules/memory-drain.md, verifying each claim against the current tree first, then clears drained rows and ships the repo edits as a PR. Use when the user says "drain memory", "triage memory", the SessionStart nudge fires, or the inbox has grown stale. Read-write — edits repo docs + deletes local memory files.
EnglishThe deterministic VCS procedures of an end-of-session git cleanup — path resolution, pre-flight audit, the poll-until-stable mergeability helper, pre-flight cross-checks (worktree/salvage/lock), per-PR squash-merge + branch-delete mechanics, the protected-branch guards, the stale-branch sweep, diverged-branch recovery, bringing develop to latest, the regression-gate build commands, the orphan-scenario sweep, and the final-report template. Invoked by git-janitor (which keeps the judgment: hard refusals, merge-gate orchestration, the FF-clean decision). Use when running post-merge cleanup, sweeping stale branches, or bringing develop to latest.
EnglishScenario-aware PR-time perf review. Given a PR number or diff path, classifies touched files via the curated diff→scenario map, runs the affected subset via scripts/dev/perf-run.sh, compares against baselines via scripts/dev/perf-compare.py, posts a delta markdown table. Use when a PR is near merge and the orchestrator wants targeted perf coverage without running the full 14-scenario suite.
EnglishAdd or remove SMATCHETUIPERFSCOPE("perftemp:...") markers per a spec provided by perf-detective or spike-hunter. Encodes the overhead rules (string-literal scope names, no nesting in million-call loops, one outer scope always safe, mandatory perftemp: prefix, header include check). Use for inserting instrumentation or stripping all perftemp: markers after a perf round.
EnglishRecall facts from prior session scratchpads — list archives, show a specific archive, or grep across all archives. Use when the user references "last session", "yesterday's run", "what did find earlier", or any cross-session continuity question. Archives live at .session-context.archive/ and are written by agents/scripts/core/clear-session-context.sh on SessionStart. Read-only — never writes.
EnglishIncludes Agent Tokens, Author Plan Doc, Coderabbit Handoff, and other workflows.
Supported by the Atlas install prompt. Skill-specific requirements vary.
Pack stats
What’s included
Agent Tokens
Report per-agent token usage for the current Claude Code session (default), all sessions (--all), or a time window (--since {h|d|w}). Reads .claude/.agent-tokens.jsonl produced by the SubagentStop hook and prints a fixed-width table with dollar estimates per Anthropic public pricing.
Author Plan Doc
Author a new plan-doc at docs/plans/active/ .md from a one-line request — copy the template, set the slug to the basename, fill every section (N/A — , never delete), include the Perf-review-system-gates section when the diff would touch the project's perf-gated core source tree, keep the three mandatory Verification/Out-of-scope items the template bakes in (doc-validation suite, grill-with-docs stress-test, deferral residue-sweep), then git add + wip(plan) commit immediately (working-tree-only plans are lost on checkout). Use when the user says "write a plan", "draft a plan", "new plan", "plan for X", or "plan this". The authoring counterpart to grill-with-docs (the stress-test). Read-write — creates + commits the plan file.
Coderabbit Handoff
The deterministic report-shape mechanics of a CodeRabbit-triage round — the triage-table example, the per-finding N handoff-packet skeleton (VALID + REJECTED worked examples), and the reply-to-bot lines. Invoked by coderabbit-triage (which keeps the judgment: the 19-rule override table, path→agent routing, severity classification, watcher-invocation, hand-back contract). Use when formatting a CodeRabbit triage report or a per-finding handoff packet.
Debug Instrument
The deterministic mechanics of a Smatchet debug investigation — the NDJSON [temp-debug] instrumentation recipe, build + exe-staleness commands, the unified-CLI run reference, log/evidence reading (jq), sanitizer setup, cleanup, and the two report-shape templates. Invoked by debug-detective (which keeps the judgment: hypotheses, metric choice, the wait-for-feedback loop, hand-off). Use when instrumenting, reading a debug log, cleaning up [temp-debug] markers, or formatting a debug report.
Drain Memory
Drain the harness auto-memory inbox into the repo. Triages every memory file under ~/.claude/projects/ /memory/ (except MEMORY.md) into implement / backlog / toss per docs/agent-rules/memory-drain.md, verifying each claim against the current tree first, then clears drained rows and ships the repo edits as a PR. Use when the user says "drain memory", "triage memory", the SessionStart nudge fires, or the inbox has grown stale. Read-write — edits repo docs + deletes local memory files.
Git Cleanup Procedures
The deterministic VCS procedures of an end-of-session git cleanup — path resolution, pre-flight audit, the poll-until-stable mergeability helper, pre-flight cross-checks (worktree/salvage/lock), per-PR squash-merge + branch-delete mechanics, the protected-branch guards, the stale-branch sweep, diverged-branch recovery, bringing develop to latest, the regression-gate build commands, the orphan-scenario sweep, and the final-report template. Invoked by git-janitor (which keeps the judgment: hard refusals, merge-gate orchestration, the FF-clean decision). Use when running post-merge cleanup, sweeping stale branches, or bringing develop to latest.
Perf Gatekeeper
Scenario-aware PR-time perf review. Given a PR number or diff path, classifies touched files via the curated diff→scenario map, runs the affected subset via scripts/dev/perf-run.sh, compares against baselines via scripts/dev/perf-compare.py, posts a delta markdown table. Use when a PR is near merge and the orchestrator wants targeted perf coverage without running the full 14-scenario suite.
Perf Instrument
Add or remove SMATCHETUIPERFSCOPE("perftemp:...") markers per a spec provided by perf-detective or spike-hunter. Encodes the overhead rules (string-literal scope names, no nesting in million-call loops, one outer scope always safe, mandatory perftemp: prefix, header include check). Use for inserting instrumentation or stripping all perftemp: markers after a perf round.
Scratchpad Recall
Recall facts from prior session scratchpads — list archives, show a specific archive, or grep across all archives. Use when the user references "last session", "yesterday's run", "what did find earlier", or any cross-session continuity question. Archives live at .session-context.archive/ and are written by agents/scripts/core/clear-session-context.sh on SessionStart. Read-only — never writes.
Quick start
- Choose your skills. Select individual skills above, or choose the entire pack.
- Copy the install prompt. Use Copy install prompt above the list.
- Send it to your agent. Paste into your coding agent’s chat. Your agent reviews the source and guides the setup.
Try it with your agent
After installation, start with a request like:
Use the Agent Tokens skill to help with my task.
Ask me for the context you need before getting started.Requirements & license
- A coding agent that can access your project files and install skills.
- Your agent checks each skill’s instructions, dependencies, and permissions during setup.
- Reported license for the skills shown: MIT. The author’s terms apply.