skills.rest
Neko-Catpital-Labs
Neko-Catpital-LabsOFFICIAL
Explore all skills by this author →
GitHub

worker-session-mine

Mines terminal agent sessions for mechanical thrash and files follow-up tasks.

18|5|Updated Apr 17, 2026
Search Tags:#token usage#session mining#agent sessions#invoker#worker automation#thrash detection#follow-up routing
One-click install
npx skills add https://github.com/Neko-Catpital-Labs/Invoker --skill worker-session-mine-neko-catpital-labs
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: worker-session-mine
Source: https://github.com/Neko-Catpital-Labs/Invoker/tree/main/skills/worker-session-mine
Command: npx skills add https://github.com/Neko-Catpital-Labs/Invoker --skill worker-session-mine-neko-catpital-labs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Fire-and-forget agent sessions (Claude, Codex, OMP) often waste tokens on mechanical thrash — repeated commands, excessive turns, and runaway token consumption — without anyone noticing. This Skill detects that waste and files a follow-up so the underlying prompts or skills can be fixed. ## Core Features & Use Cases - Multi-harness session discovery: Resolves Claude, Codex, and OMP sessions via Invoker task inventory using agentSessionId and agentName. - Thrash detection: Flags sessions exceeding thresholds such as 40+ assistant turns, 10M+ total tokens, or the same bash command repeated 5+ times. - Routed follow-ups: Files findings to catstack for skill prose issues or to Invoker for harness prompt issues, with strict rate caps (one follow-up per session hash per week). - Use Case: An operator enables the worker on a DO1 owner, and it periodically scans completed agent sessions, filing a follow-up PR proposal when a session burned 12M tokens repeating the same failing command. ## Quick Start Enable the worker-session-mine desired state from the Workers UI or run invoker-cli worker toggles --enable worker-session-mine on the DO1 owner.

Frequently Asked Questions about worker-session-mine

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I enable the worker-session-mine worker in Invoker?▼

Enable it through the Workers UI by toggling desired state, or run invoker-cli worker toggles --enable worker-session-mine. It should only be enabled on the DO1 owner, not on Mac owners by default.

What counts as thrash in an agent session?▼

Thrash is flagged when a session has 40 or more assistant or Codex turns, 10M or more cache-read tokens, 10M or more total tokens, or the same bash argv repeated 5 or more times. The catstack token_audit.py script can also flag sessions when CATSTACK_ROOT is set.

Which agent harnesses does session mining support?▼

It supports Claude, Codex, and OMP sessions. Claude sessions resolve via CLAUDE_CONFIG_DIR or ~/.invoker/claude-worker/projects, Codex via ~/.invoker/agent-sessions/*.jsonl, and OMP via *.omp.txt files.

Where do follow-up findings get filed?▼

Findings about skill prose, methodology, or hooks go to catstack PRs only. Findings about the Invoker harness prompt or product go to Invoker PRs only. The worker never merges follow-up PRs itself.

What are the rate limits for follow-up filings?▼

The worker caps output at one follow-up per session hash per week, a maximum of one per tick, and two per day. This prevents duplicate or noisy follow-up filings.

Can worker-session-mine be used for interactive reflection?▼

No. Interactive /reflect stays in catstack and must not be vendored into this repo. This worker only handles off-by-default mining of terminal fire-and-forget sessions.

Related Skills

Explore More Software Engineering Skills→
golemcloudgolemcloud
1.5k

golem-invoke-agent-scala

Invokes methods on deployed Scala Golem agents from the command line.

Official
Intermediate
auldsyababuaauldsyababua
6

prompt-optimization-analyzer

Analyze skill prompts for token waste, anti-patterns, and trigger issues.

Community
Basic
nextcnextc
2

context-budget

Audit Claude Code context window usage across agents, skills, MCP servers, and rules.

Official
Advanced
AMindToThinkAMindToThink
4

cookie

Provide a literary reflective passage for end-of-session pauses.

Community
Basic