psychology-agent: Skill for Claude Code

Doc — Mid-Work Documentation Persistence

Capture what just happened — decisions, findings, design choices, reasoning — and write it to the correct file on disk before context compression loses it.

Design principle: /doc is the complement of /cycle. Where /cycle is post-hoc ("we're done, update everything"), /doc is mid-work ("something worth remembering just happened — save it now"). Together they ensure nothing falls through the gap between discovery and documentation.

When to trigger:

• A design decision was just made or resolved
• A finding emerged from analysis or discussion
• An architectural choice was debated and settled
• Context usage is above 50% and substantive work hasn't been persisted
• The agent notices it's holding reasoning that only exists in conversation

When NOT to trigger:

• Trivial exchanges (greetings, clarifications, small corrections)
• Work that /cycle will handle at session end
• Information already on disk (check before writing)

Phase 1: Identify What to Document

Parse $ARGUMENTS and recent conversation context to determine:

1. What happened — the decision, finding, or reasoning to persist
2. Type classification:

 Type              Examples                           Typical destination
──────────────────────────────────────────────────────────────────────────
 Design decision   "evaluator uses tiered activation"  MEMORY.md, CLAUDE.md,
                                                       or architecture doc

 Finding           "skills can't natively chain"        MEMORY.md or
                                                        research doc

 Reasoning chain   "why PJE is a case study not a       Snapshot file or
                    sub-agent"                           architecture doc

 Convention        "always use LaTeX for complex docs"   MEMORY.md

 Architecture      "four-layer agent system"              Architecture doc

 Status update     "PSQ analysis complete, moving to     MEMORY.md
                    architecture"