ce-compound-refresh

Reviews and refreshes stale learning docs in docs/solutions/ against the current codebase.

Updated May 11, 2026
One-click install
npx skills add https://github.com/mmnavarr/harness --skill ce-compound-refresh-mmnavarr
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ce-compound-refresh
Source: https://github.com/mmnavarr/harness/tree/main/skills/ce-compound-refresh
Command: npx skills add https://github.com/mmnavarr/harness --skill ce-compound-refresh-mmnavarr

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Learning and pattern docs under docs/solutions/ drift out of sync with the codebase over time, leaving stale references, contradictions, and redundant documents that mislead future work. This Skill audits those docs against current code and keeps the knowledge base accurate. ## Core Features & Use Cases - Drift Detection and Classification: Cross-references each learning doc against the current codebase and classifies it as Keep, Update, Consolidate, Replace, or Delete based on evidence. - Document-Set Analysis: Detects overlapping, superseded, and contradictory docs, then consolidates them into canonical sources of truth. - Interactive and Headless Modes: Runs interactively with one-at-a-time user questions, or fully unattended via mode:headless with an applied/recommended summary report. - Use Case: After months of refactoring, run the refresh to find learnings referencing deleted files, merge two overlapping auth docs into one canonical doc, and delete obsolete guidance while git history preserves it. ## Quick Start Ask the AI to refresh my learnings in docs/solutions/ and clean up any stale or overlapping docs.

Frequently Asked Questions about ce-compound-refresh

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

FAQPage Schema
How do I refresh stale documentation against my codebase?▼

Invoke the refresh workflow on your docs/solutions/ directory. It reads each learning doc, cross-references file paths, class names, and code snippets against the current code, then classifies each doc as Keep, Update, Consolidate, Replace, or Delete with evidence.

How do I run the doc refresh without answering questions?▼

Add mode:headless to the invocation arguments. Headless mode skips all user questions, applies unambiguous actions automatically, marks ambiguous cases as stale, and produces a summary report split into applied and recommended actions.

Can I limit the refresh to one module or directory?▼

Yes, pass a scope hint as an argument such as a subdirectory name, filename, module, or keyword. The skill matches it against directories, frontmatter fields, filenames, then content, stopping at the first strategy that produces results.

What happens to deleted learning docs?▼

Deleted docs are removed outright with no archive directory. Git history preserves every deleted file, so you can recover them later with git log --diff-filter=D on the docs/solutions/ path.

When does the skill replace a doc instead of updating it?▼

Replace applies when the core recommended solution conflicts with current code, not just when references moved. A replacement subagent writes a successor doc following the schema and template, then the orchestrator deletes the old file.

Does the refresh check for links to a doc before deleting it?▼

Yes, it searches the repo's markdown content for citations of the file before deletion. Decorative citations allow deletion with cleanup, while substantive citations signal Replace or stale-marking instead.