Itsdevcoffee Devcoffee Agent Skills Feedback

feedback

Evaluate TLDR command outputs against quality criteria and log results for improvement tracking. This is a development tool for improving the /tldr skill itself.

Path Resolution

All evaluation data lives inside the TLDR plugin directory. This skill is at skills/feedback/SKILL.md — the plugin root is two directories up (the directory containing skills/, docs/, and .claude-plugin/). All file paths below are relative to the plugin root.

• Scoring rubrics: skills/feedback/references/EVALUATION.md
• Evaluation log: docs/evaluation/evaluation-log.md
• Sample files: docs/evaluation/samples/NNN-description.md
• Notes catalog: docs/evaluation/notes.md

Evaluation Workflow

Step 1: Identify the Sample

Determine the input mode:

• Interactive (default): User pastes or references an original message and its TLDR output
• Automated (--no-user-score): Called from /tldr:note --ex — skip user prompts, agent scores only

Step 2: Load Scoring Rubrics

Read references/EVALUATION.md (bundled with this skill) for the complete scoring criteria, detailed rubrics for each dimension, and the sample file format template.

Step 3: Determine Next Sample ID

Read docs/evaluation/evaluation-log.md and find the highest existing sample ID. The new sample is that number + 1, zero-padded to 3 digits (e.g., 004).

Step 4: Score the Sample

Evaluate on four criteria (Completeness, Conciseness, Actionability, Accuracy) at 2.5 points each — detailed rubrics are in references/EVALUATION.md.

For each criterion, produce: (1) items captured correctly, (2) items missed or needing improvement, (3) a numeric score with one-line justification.

Step 5: Provide Analysis

Include in the evaluation:

• What Went Well — Specific strengths
• What Needs Work — Specific weaknesses
• Recommendations — Actionable improvements for future TLDR versions

Step 6: Create Sample File

Write to docs/evaluation/samples/NNN-description.md following the format template in references/EVALUATION.md:

• Header with date, type, context
• Original message (or URL reference if external)
• TLDR output
• Claude's evaluation with breakdown
• User's evaluation section (populated or "Not provided")

Step 7: Update Evaluation Log

Update docs/evaluation/evaluation-log.md:

1. Increment "Samples Collected" count
2. Recalculate all averages (overall + per-criterion)
3. Add row to the evaluation samples table
4. Update "Samples with User Feedback" count if user scored

Step 8: User Score Handling

Interactive mode (default):

1. Present the agent's score and analysis
2. Ask the user for their score (0.0-10.0) and optional feedback
3. If user provides a score, update the sample file and evaluation log

Automated mode (--no-user-score):

• Skip user prompts entirely
• Mark sample status as "Unscored" in the log
• Set user evaluation to "Not provided" / "None"

Output

After logging, provide a brief summary in chat:

Sample 004 logged (8.5 / 10.0)
- Strengths: Captured all key findings and file paths accurately
- Needs work: Two bullets could be more action-oriented
- Running tally: 4 samples, 9.1 avg, 1 with user feedback

Saved to docs/evaluation/samples/004-auth-refactor.md

Converted and distributed by TomeVault — claim your Tome and manage your conversions.