Claude-Mind: Skill for Claude Code

/reflect-on-outcome — Outcome-Based Reflection

This sub-skill implements all outcome-based reflection modes for /reflect. It is invoked by the parent router for three modes:

• Hypothesis mode (--on-hypothesis <id>): Full single-hypothesis reflection pipeline
• Execution mode (--on-execution): Lightweight execution outcome reflection
• Batch micro mode (--batch-micro): Batch micro-hypothesis processing

Each mode section below is self-contained with its own step numbering.

Mode: Hypothesis Reflection (--on-hypothesis )

This sub-skill implements Mode 1 of /reflect. It is invoked by the parent /reflect router when --on-hypothesis <hypothesis-id> is specified, or during --full-cycle for each newly resolved hypothesis. It runs the full reflection pipeline: horizon gate, ABC chain analysis, differentiated and contrastive extraction, encoding score, textual reflection, violation tracking, source tracking, journal entry, accuracy reflection, pattern signatures, entity extraction, belief registry, contradiction detection, process-outcome classification, context gap analysis, strategy tracking, experiential index updates, spark check, knowledge reconciliation, and tree growth trigger.

Step 0: Load Conventions

Step 0: Load Conventions — Bash: load-conventions.sh with each name from the conventions: front matter. Read only the paths returned (files not yet in context). If output is empty, all conventions already loaded — proceed to next step.

Step 0.5: Horizon Gate

Read the resolved record and check its horizon field (default: "short" if missing).

IF horizon == "micro":
    ERROR: Micro-hypotheses should never reach --on-hypothesis.
    They are batch-reflected via --batch-micro. Abort.

IF horizon == "session":
    Run LIGHTWEIGHT reflection path:
      - Generate ABC chain (Step 2) — abbreviated: skip skill_chain and research_depth
      - Skip Step 2.5 differentiated extraction UNLESS surprise >= 7 — EXCEPT when
        outcome == CORRECTED, where Step 2.5 runs regardless of surprise (its CORRECTED
        branch only; the CONFIRMED branch stays skipped). See "Why CORRECTED is exempt"
        below — do NOT read this as a general fall-through (guard-137).
      - Skip Step 2.6 contrastive extraction
      - Generate encoding score (Step 2.7) — normal
      - Generate textual reflection (Step 3) — abbreviated (3-5 sentences max)
      - Skip Step 3.5 domain-specific reflection
      - Track violation (Step 4) — if corrected, normal
      - Update source tracking (Step 5) — only if external sources used
      - Append to journal (Step 6) — abbreviated entry
      - Accuracy reflection (Step 7) — normal
      - Skip Steps 7.5, 7.5b, 7.6, 7.6b (pattern sigs, entities, beliefs, contradictions)
      - Skip Step 7.6c process-outcome dual score
      - Skip Step 7.7 context gap analysis
      - Run spark check (Step 8) — normal (short hypotheses can still spark insights)
    RETURN after lightweight path — do not continue to full Step 1.

    # Meta-strategy category depth override
    Bash: meta-read.sh reflection-strategy.yaml
    IF hypothesis.category in category_depth_overrides:
        override = category_depth_overrides[hypothesis.category]
        IF override == "full" AND horizon == "session":
            Log: "META OVERRIDE: session horizon → full pipeline for category {hypothesis.category}"
            CONTINUE to full reflection pipeline (skip lightweight return)

IF horizon == "short" OR horizon == "long" OR horizon is missing:
    Continue to full reflection below.