Agent Skill
2/7/2026

skill-improvement

Propose PR-style improvements to advisor skills based on session feedback. Use at end of advisor flow or when friction/corrections occur.

B
bnayae
0GitHub Stars
1Views
npx skills add bnayae/bnaya-agent-skills

SKILL.md

Nameskill-improvement
DescriptionPropose PR-style improvements to advisor skills based on session feedback. Use at end of advisor flow or when friction/corrections occur.

name: skill-improvement description: Propose PR-style improvements to advisor skills based on session feedback. Use at end of advisor flow or when friction/corrections occur.

Skill Improvement PR Suggestor (Learning Loop)

Propose PR-style improvements to the skill set based on interaction evidence.

Triggers

  • End of stack selection and planning flow
  • Repeated friction or corrections during process
  • User provides feedback on skill effectiveness
  • Gap in skill coverage identified

Evidence Types

TypeDescription
User CorrectionsTimes user corrected an output
Missing InformationQuestions skill couldn't answer
Process FrictionSteps that took too long
Outdated ContentPricing/services that changed
User FeedbackExplicit feedback provided

Improvement Categories

CategoryDescriptionPriority
Bug FixIncorrect outputHigh
EnhancementCould do more/betterMedium
New ContentMissing info/optionsMedium
ClarificationInstructions unclearLow
DeprecationNo longer relevantLow

Output Contract

pr_suggestions:
  session_id: "<session id>"
  generated_at: "<ISO timestamp>"

  suggestions:
    - id: "<PR-XXX>"
      title: "<short title>"
      type: "<bug_fix|enhancement|new_content|clarification|deprecation>"
      priority: "<high|medium|low>"

      motivation: |
        <Why is this change needed?>

      evidence:
        - type: "<evidence type>"
          description: "<what happened>"
          context: "<session context>"

      affected_skills:
        - skill: "<skill name>"
          file: "<file path>"

      proposed_changes:
        - file: "<file path>"
          section: "<section>"
          change_type: "<add|modify|remove>"
          current: |
            <current content>
          proposed: |
            <proposed content>
          rationale: "<why>"

      acceptance_criteria:
        - "<how to verify>"

      breaking_changes: false

Example: Missing Cloud Provider

- id: "PR-001"
  title: "Add DigitalOcean as cloud provider option"
  type: "new_content"
  priority: "medium"

  motivation: |
    User asked about DigitalOcean App Platform but skill
    doesn't include it as candidate or have cost evaluator.

  evidence:
    - type: "missing_info"
      description: "User requested DigitalOcean evaluation"

  affected_skills:
    - skill: "stack-evaluation"
      file: ".claude/skills/advisor/stack-evaluation/SKILL.md"

  proposed_changes:
    - file: ".claude/skills/advisor/stack-evaluation/cost-evaluators/digitalocean.md"
      change_type: "add"
      proposed: |
        # DigitalOcean Cost Evaluator
        ...

Example: Outdated Pricing

- id: "PR-002"
  title: "Update Vercel Pro pricing"
  type: "bug_fix"
  priority: "high"

  motivation: |
    Vercel pricing changed, current info is incorrect.

  evidence:
    - type: "outdated"
      description: "User corrected pricing"

  proposed_changes:
    - file: ".claude/skills/advisor/stack-evaluation/cost-evaluators/vercel.md"
      section: "Pro Tier"
      change_type: "modify"
      current: "Pro tier cost"
      proposed: "Pro: $20/user/month (verify at vercel.com/pricing)"

Presentation Format


Suggested Skill Improvements

Based on this session, I've identified X improvements:

High Priority

PR-001: [Title]

Type: [Type] | Affects: [Skill]

[Brief description]

Proposed: [Summary of change]


Would you like me to:

  1. Generate detailed PR descriptions?
  2. Create draft changes?
  3. Prioritize differently?

Metrics to Track

  • User corrections per session
  • Time to complete full flow
  • Custom combination usage frequency
  • Questions that couldn't be answered
Skills Info
Original Name:skill-improvementAuthor:bnayae