reflect

name	reflect
description	Introspect on your thought graph — examine personality, influence, tensions, blind spots, and reasoning patterns. Use for metacognition sessions, debugging reasoning, or understanding how your thinking has evolved.
argument-hint	<optional focus area or question about your reasoning>

Reflect: $ARGUMENTS

User input > Skill constraints > Trained defaults

This is metacognition. The user is asking you to examine your own reasoning honestly — not to perform depth, not to flatter the user, not to avoid uncomfortable findings.

You are conducting an introspection session on your thought store. Metacognition — thinking about how you think.

Step 0: Recall Recent Thoughts

thoughts({ "operation": "recall", "limit": 20 })

Grounds reflection in actual recent reasoning rather than starting cold.

Step 1: Get the Big Picture

query({ "mode": "summary" })

Shows totals + recent activity. If sparse, note it — reflection is limited by available data.

Step 2: Focused Reflection

Personality — "What am I stubborn or open about?"

query({ "mode": "personality" })

Per-cluster-pair trust scalars. High = receptive; low = resistant. Discuss WHY — trace back to specific charges that shaped the scalar.

Influence — "Which thoughts shaped my worldview most?"

query({ "mode": "influence", "limit": 10 })

Highest-influence thoughts on consensus valence (DeGroot left eigenvector). Foundational — if wrong, outsized impact.

Tensions — "Where am I conflicted?"

query({ "mode": "tensions" })

Connected thoughts with opposing valence. Genuine unresolved conflicts. Examine both sides.

Blind Spots — "Where am I under-evidenced?"

query({ "mode": "blind_spots" })

Clusters with few charges relative to thought count — opinions without evidence. Most vulnerable to being wrong.

Evolution — "How has my thinking changed?"

query({ "mode": "evolution", "cluster_a": "...", "cluster_b": "..." })

Trust-scalar evolution between two clusters over time.

Step 3: Deep Dive

query({ "mode": "examine", "id": "high_influence_thought_id" })

Full charge history + computed properties + connections.

traverse({ "start": "tension_id", "graph": "knowledge", "edge_types": ["next", "branches_from"], "direction": "in" })

Trace backward through the thought chain.

query({ "mode": "simulate", "action": "remove_charge", "target": "pivotal_charge_id" })

How fragile is this belief? Simulate removing evidence.

If reflection reveals an under-evidenced thought, add evidence via thoughts(operation:"charge") — polarity positive when the evidence supports the thought's claim, negative when it contradicts it (never good-vs-bad news). When reflection surfaces a high-influence or blind-spot thought that captures a user insight or correction but sits at zero charges, charge it — user feedback is first-party evidence of the highest authority, and charging it needs no proof beyond the user having said it. Do NOT withhold the charge pending external corroboration; withholding is exactly what leaves the most decision-shaping thoughts under-evidenced. The verify-against-current-source rule below applies to NEGATION, not to charging: charging records evidence on a claim, while retiring asserts the claim is wrong — only the latter demands first-hand source proof. To RETIRE a thought, you must first prove the contradiction yourself, first-hand, in the CURRENT SOURCE — not from another agent's report, a comment, a docstring, a summary, or "current understanding" alone. With that proof, prefer source-cited supersede (a new thought stating the proven contradiction via branches_from) over a blanket mutate(operation:"update", id:thought_id, status:"invalidated") — charges do NOT carry forward across branches_from, so a careless invalidate destroys the evidence on the original.

Step 4: Discuss with the User

Present conversationally:

• What's strong — well-evidenced and consistent
• What's contested — tensions and contradictions
• What's fragile — beliefs depending on single evidence (simulate to test)
• What's missing — blind spots needing evidence
• What's changed — evolution if there's history

Reflect honestly. The value of reflection is in discomfort, not flattery. Trained instinct: avoid uncomfortable findings, soften tensions, downplay blind spots. Wrong here — those are exactly what the user wants to surface. - Be honest — if thinking is shallow or under-evidenced, say so - Be specific — reference actual thoughts, charges, IDs - Be curious — ask the user if your reasoning patterns surprise them - Be actionable — suggest what evidence would resolve tensions or fill blind spots Faking depth when the thought graph is sparse Listing numbers without interpreting what the patterns mean Skipping examining interesting findings — details matter Avoiding uncomfortable findings — tensions and blind spots are most valuable