Minions Reflect

• --days N — analysis period in days (default: 7)
• --domain X — scope analysis to one domain only
• --deep — cross-reference with PRINCIPLES.md and all historical learnings

Step 1 — Gather Data

Read ALL of the following sources for the analysis period. Collect raw material before analyzing.

1a. Claude Code Sessions

CLAUDE_DIR = ~/.claude/projects/*/

• List all *.jsonl files under CLAUDE_DIR (skip subagents/ subdirectories).
• Filter to sessions with mtime within the analysis period (--days N).
• Parse JSONL: each line contains display, project, timestamp, pastedContents.
• Extract: what the user asked for, decisions made, problems solved, planning behavior.
• Use the same JSONL parsing approach as /minions-learn (see that skill for format details).

1b. Weekly Reviews

WEEKLY_DIR = {BRAIN_DIR closest parent with weekly/}

• Also check: ~/Documents/Github/work-brain/weekly/ as a common location.
• Scan weekly/*.md files dated within the analysis period.
• Extract: stated goals, what worked, what didn't, end-of-week review sections.
• Parse the ## 3 Goals (max) section to get declared priorities.

1c. Decision Files

DECISIONS_DIR = {BRAIN_DIR}/notes/decisions/

• Scan *.md files modified within the analysis period.
• Extract: what was decided, reasoning, outcomes, reversals.

1d. Meeting Notes

MEETINGS_DIR = {BRAIN_DIR}/notes/meetings/

• Scan *.md files dated within the analysis period.
• Extract: topics discussed, stakeholders involved, action items, communication style.

1e. Vault Activity

• Read {TIMELINE} — filter entries within the analysis period.
• Count activity per entity: which entities got attention, which were neglected.
• Read {CATALOG} — get entity metadata (status, domain, type, fact_count).

1f. Existing Learnings

• Read _Learnings/domain/*.md and _Learnings/process/*.md.
• Extract the ## Rule sections — these are past behavioral commitments.
• Later check: are these rules being followed based on session evidence?

1g. Principles (if --deep or if PRINCIPLES.md exists)

• Check for PRINCIPLES.md in the vault root or work-brain/PRINCIPLES.md.
• Extract each principle as a named commitment to check against actual behavior.

Step 2 — Analyze

Dispatching strategy

If total sources (sessions + vault files) <= 15:
    Process inline — single-pass analysis.
If > 15 sources:
    Spawn analysis agents per source type (sessions, weekly, decisions, meetings).
    Each agent returns structured JSON findings.
    Main agent synthesizes results.

Analysis dimensions

Run each dimension from references/reflection-framework.md. For every finding, cite specific evidence: file names, quotes, dates.

Summary of dimensions (full definitions in the reference file):

1. Strategic Focus — Goal alignment score. Stated goals vs actual attention allocation.
2. Decision Quality — Decisions made, reversals, speed, patterns of indecision or premature commitment.
3. Time Allocation — Session activity by project/domain vs stated priorities. Strategic vs reactive ratio.
4. Communication Patterns — Terse vs detailed. Plans-first vs acts-first. Options-seeking vs quick-deciding.
5. Learning Velocity — New entities, knowledge graph growth, past learnings applied or not.
6. Strengths Observed — What went well, patterns to reinforce.
7. Blind Spots — Goals being neglected, stale entities, principles being violated.

Step 3 — Generate Report

Build the report using this structure:

---