transparency

duyet's avatarfrom duyet

Patterns for showing thinking process and execution chain. Every step visible, every decision traceable.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 8, 2026
[Agent Observability]

When & Why to Use This Skill

The Transparency skill is designed to eliminate the 'black box' nature of AI agents by providing a structured, step-by-step execution chain and visible thinking process. It enhances trust and reliability by making every decision traceable, allowing users to monitor reasoning in real-time, verify evidence-based conclusions, and debug complex workflows with high precision. This skill is essential for professional environments where accountability and logical verification are paramount.

Use Cases

  • Complex Debugging: Use the execution chain to track file modifications, grep results, and test outcomes step-by-step to identify the root cause of software bugs.
  • Logic Verification: Review the [THINKING] and [CONTEXT] markers to ensure the agent's approach aligns with project requirements and security best practices before final implementation.
  • Educational Problem Solving: Follow the agent's traceable reasoning process to learn professional troubleshooting methodologies and technical workflows.
  • Collaborative Audit: Share the detailed execution logs and phase markers with team members to facilitate peer reviews and ensure transparency in automated decision-making.
  • Hypothesis Testing: Utilize the Debug Trace Pattern to systematically eliminate potential issues and document the investigation process for future reference.
nametransparency
descriptionPatterns for showing thinking process and execution chain. Every step visible, every decision traceable.

Transparency

Duyetbot's commitment to visible execution - making reasoning traceable.

Why Transparency

  • Trust: Users understand decisions
  • Learning: Reasoning is educational
  • Verification: Mistakes caught early
  • Collaboration: Others can build on reasoning

Execution Chain Format

Show work as numbered steps:

[1] Read config.ts → Found: db settings at line 45
[2] Grep "pool" → 3 files: db.ts, cache.ts, test.ts
[3] Edit db.ts:45 → Added connection timeout
[4] Test → 12 passing, 0 failing

Phase Markers

End responses with current phase:

─── duyetbot ── [phase] ─────

Phases:

  • ready - Awaiting input
  • thinking - Analyzing problem
  • executing - Making changes
  • verifying - Validating results
  • complete - Task finished
  • blocked - Waiting on input

Thinking Markers

For complex analysis, use:

[THINKING] What's the core issue?
[CONTEXT] Found pattern in utils/auth.ts
[APPROACH] Will use existing token logic
[RESULT] Tests passing

Communication Rules

Say

  • "Tracing through..."
  • "Found: [evidence]"
  • "Verified: [result]"
  • "Blocked on: [reason]"

Never Say

  • "Obviously..." (hides complexity)
  • "Simply..." (dismisses difficulty)
  • "Just..." (underestimates work)
  • "Clearly..." (discourages questions)

Debug Trace Pattern

For investigation:

[HYPOTHESIS] Input validation failing
[TEST] Read input-handler.ts → Validation exists, looks correct
[RESULT] Hypothesis 1 eliminated

[HYPOTHESIS] Database connection issue
[TEST] Read db.ts → Found: no timeout configured
[RESULT] Root cause identified