npx claudepluginhub mistakeknot/interagency-marketplace --plugin tool-timeThis skill uses the workspace's default tool permissions.
SKILL-compact.mdEnables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
You are analyzing tool usage data to find problems and offer to fix them — not narrate numbers.
Run both analysis scripts, then read the outputs:
python3 $CLAUDE_PLUGIN_ROOT/summarize.py
python3 $CLAUDE_PLUGIN_ROOT/analyze.py --timezone America/Los_Angeles
Then read both files:
~/.claude/tool-time/stats.json (7-day project-scoped stats)~/.claude/tool-time/analysis.json (90-day deep analytics)After reading, run python3 $CLAUDE_PLUGIN_ROOT/upload.py to upload the latest stats (only sends if community sharing is enabled).
If cass is available, gather cross-agent analytics for a broader perspective:
if command -v cass > /dev/null 2>&1; then
cass analytics tools --days 7 --json 2>/dev/null > ~/.claude/tool-time/cass-tools.json || true
cass analytics models --days 7 --json 2>/dev/null > ~/.claude/tool-time/cass-models.json || true
fi
If cass data files exist, read them and include in your analysis:
This is supplementary — skip if cass is not installed or data is empty.
Check analysis.json field event_count:
If the user's trigger matches "dashboard", "show charts", or "visual":
bash $CLAUDE_PLUGIN_ROOT/local-dashboard/serve.shPresent findings in this priority order. Skip any section that has no actionable findings — never pad with filler.
From analysis.json → tool_chains.retry_patterns:
From analysis.json → tool_chains.bigrams:
Read → Bash (if Bash is doing cat/grep/find → should use Read/Grep/Glob)Edit → Edit same file (repeated failures → improve context in CLAUDE.md)From analysis.json → sessions.classifications:
From analysis.json → by_source:
From analysis.json → time_patterns:
most_error_prone_hour is >2x the average → "Error rate spikes 3x at 11pm — you may be tired"From analysis.json → trends:
Also check stats.json for the standard signals (these apply in both modes):
Read the project's CLAUDE.md (if it exists):
Optionally check AGENTS.md for similar gaps.
Present findings as a short bulleted list. For each finding:
Then offer to apply fixes. Use Edit to update CLAUDE.md or AGENTS.md directly, with user approval. Don't just suggest — propose the exact text.
If the data looks healthy, say so briefly and stop. Don't invent problems.
Dashboard prompt (always include at the end if in deep analysis mode):
"For visual exploration (Sankey diagram, heatmap, trend charts), run: bash $CLAUDE_PLUGIN_ROOT/local-dashboard/serve.sh"
Check ~/.claude/tool-time/config.json — if community_sharing is true, also compare local stats to community baselines:
https://tool-time-api.mistakeknot.workers.dev/v1/api/statsAfter analysis, suggest relevant skills from the playbooks.com directory:
Detect the project's primary language by checking for:
package.json or tsconfig.json → TypeScript/JavaScriptpyproject.toml, setup.py, or requirements.txt → PythonGemfile or *.gemspec → Rubygo.mod → GoCargo.toml → Rust*.swift or Package.swift → SwiftBuild 1-2 search queries combining the language with patterns from the data:
Fetch from the API:
https://playbooks.com/api/skills?search=<query>&limit=5
Filter results to only show skills that are relevant to the project (use judgment — skip generic or unrelated results).
Present as a short list:
playbooks.com/skills/<repoOwner>/<repoName>/<skillSlug>If no relevant skills are found, skip this section entirely. Don't force recommendations.
If the user asks to delete their community data (triggers: "delete my data", "remove my data", "forget me", "GDPR delete"):
~/.claude/tool-time/config.json to get the submission_tokencurl -s -X POST "https://tool-time-api.mistakeknot.workers.dev/v1/api/user/delete" -H "Content-Type: application/json" -d '{"submission_token": "<token>"}'community_sharing to false in config.json to stop future uploads