Reflection Logic

Evaluation Framework:

1. Core Thesis: What is the central argument or requirement?
2. Structural Integrity: Is the document logically consistent?
3. Architectural Alignment: Does it conflict with existing PROJECT.md or AGENTS.md directives?
   • Check: Does it violate encapsulation?
   • Check: Does it contradict the System 1/System 2 split?
4. Actionable Intelligence: What specific tasks or constraints does it impose?

2. Session Reflection (The "Mirror")

Reflect on the agent's own performance during a session.

Reflection Questions:

1. Goal Achievement: Did we meet the objectives in REQUIREMENTS.md?
2. Process Adherence: Did we follow GSD (Plan -> Execute -> Verify)?
3. Tool Usage: Did we use skills effectively (e.g., LobeHub CLI)?
4. Learning: What new knowledge was acquired (e.g., new skills, patterns)?
5. Friction Points: Where did the process stall or become ambiguous?

3. Session Reporting

Generate a structured report summarizing the session.

Report Template (Session_Report.md):

• Overview: Date, Goal, Status.
• Input Analysis: Summary of sources/ and key directives.
• Architecture Updates: Changes to PROJECT.md, STATE.md, etc.
• Skills Acquired: New skills installed or created.
• Decisions: Key architectural choices made (and why).
• Next Steps: Immediate actions for the next session.

Instructions

How to Analyze Sources

1. Read all files in .sessions/<id>/sources/.
2. Identify the "Ground Truth" documents vs. "Context" documents.
3. Extract key constraints and add them to REQUIREMENTS.md (if not already there).

How to Generate a Session Report

1. Run the reflection process (Capability 2).
2. Compile findings into the Session_Report.md format.
3. Store the report in .sessions/<id>/Session_Report.md.
4. Update STATE.md with the "Latest Update" and "Completed Actions".

Scripts

• (Planned) scripts/generate_report.py: Automated generation of the report draft.