Reflection & Learning Skill

Use this skill when the user provides corrections (feedback) or when analyzing workflow patterns.

Learned Corrections

• •Workflow: Prefer git stash over git checkout to preserve uncommitted work.
• •Workflow: Run full test suite (cargo test) locally before every commit.
• •Workflow: Write tests before finalizing feature implementation.
• •Interaction: Explicitly propose destructive/irreversible actions (commit, push, overwrite) and await confirmation.
• •Interaction: Pause execution immediately upon "No!" or alarm signals.
• •Interaction: When correcting mistakes, explain the cause and proposed fix before executing it.
• •Interaction: Distinguish between generative (code) and additive (docs) tasks; verify file state before writing to docs.
• •Workflow: Avoid reading entire large files when targeted edits are possible to prevent "looping"/stalling.

Capability: Reflect (Learn from Corrections)

When the user provides a correction (e.g., "No, do it this way", "Don't use X", "Remember to Y"):

1. •Capture the Learning: Summarize the user's correction into a concise rule.
2. •Update Memory: Append this rule to the "Learned Corrections" section in THIS file (.github/skills/reflect/SKILL.md).
3. •Confirm: Tell the user you have updated the reflection memory with the new correction.

Capability: Reflect (Discover Patterns)

When the user asks to "reflect on skills" or if you notice the user repeating a complex workflow multiple times:

1. •Analyze History: Read the conversation history to identify repeated sequences of commands or requests (e.g., "build then test then run").
2. •Synthesize: Draft a new skill definition based on the observed pattern, including a name, description, and steps.
3. •Propose: Present the draft skill to the user for review.
4. •Create: Upon user approval, create the new Skill file (e.g., .github/skills/new-workflow/SKILL.md).