s1kasl-askdjna_skill

Overview

This skill captures learnings, errors, and feature requests as structured markdown entries to enable continuous improvement. It writes to a .learnings directory and provides rules for when to log, how to format entries, and when to promote broadly applicable findings into project memory. Use it to turn transient mistakes and user corrections into durable improvements and automated follow-ups.

How this skill works

When a failure, correction, missing capability, or outdated knowledge is detected, the skill appends a structured entry to one of three files: .learnings/LEARNINGS.md, .learnings/ERRORS.md, or .learnings/FEATURE_REQUESTS.md. Entries follow a templated format with ID, timestamp, priority, status, area tags, summary, details, suggested action, and metadata to support tracking and promotion. Recurring patterns are detected by matching Pattern-Key values and can be promoted into permanent project files like CLAUDE.md, AGENTS.md, TOOLS.md, or SOUL.md.

When to use it

• A command or external tool fails, throws an exception, or times out
• A user corrects the agent (e.g., “No, that’s wrong” or provides the right info)
• A user asks for a capability that isn’t implemented or accessible
• You notice your knowledge is outdated or an API behavior differs from docs
• You discover a better, repeatable approach for a recurring task
• Before starting a major task, review .learnings for related past issues

Best practices

• Log immediately while context and reproduction steps are fresh
• Be specific: include commands, inputs, environment, and reproduction steps
• Assign priority and area tags to make triage and filtering easy
• Link related entries with See Also and use Pattern-Key for dedupe
• Promote broadly applicable rules into project memory (CLAUDE.md, AGENTS.md, TOOLS.md)
• Review and resolve entries regularly; mark status changes and add resolution details

Example use cases

• Record a failing CI job output to .learnings/ERRORS.md with reproduction steps and suggested fix
• Append a correction to .learnings/LEARNINGS.md after a user supplies missing domain facts
• Log a feature request to .learnings/FEATURE_REQUESTS.md when a user asks for a new capability
• Detect recurring input-validation failures, increment Recurrence-Count, then promote a prevention rule to SOUL.md
• Run a quick review of pending high-priority learnings before a release or major refactor