Can it collect results if a subagent isn't finished?

Yes, it supports partial collection for subagents that hit context limits or require intervention, allowing developers to preserve work already completed.

What files does it use to verify completion?

It primarily checks for a .completion.json marker file, which is a lightweight alternative to parsing large session history files.

How does it handle context compaction?

It parses all assistant entries in the session file to ensure token totals are cumulative, accurately accounting for usage that occurred both before and after compaction events.

Why is token reporting mandatory in this skill?

Since users cannot observe subagent execution directly, token reporting provides the only visibility into execution quality and resource usage, helping verify if the subagent performed efficiently.

Subagent Result Collector

Name: Subagent Result Collector
Author: cowwoc

bycowwoc

•

生産性とワークフロー

Gathers work products, commit history, and precise token metrics from completed AI subagents for seamless integration.

The Collect Results skill automates the extraction of results from AI subagent worktrees, ensuring that code changes, git commits, and performance metrics are accurately captured. It provides critical visibility into subagent execution by parsing session files and completion markers, reporting precise token usage and identifying potential quality issues like context compaction. By standardizing the collection process, it prepares subagent outputs for merging into parent tasks while maintaining a clear audit trail of the agent's progress and resource consumption.

主な機能

01Lightweight verification using completion markers for high-performance status checks

02Precise token usage tracking across all context compaction events

0356 GitHub stars

04Automated extraction of git commit history and file diffs from subagent worktrees

05Standardized reporting and parent state updates for project tracking

06Identification and reporting of issues discovered during subagent execution

ユースケース

01Extracting partial progress from subagents that hit context limits or stalled

02Merging a completed subagent's code changes back into the main task branch

03Auditing AI token consumption and execution quality for complex coding tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add cowwoc/cat collect-results

For use in Claude.ai and ChatGPT

Download Skill