Heal

Subsystem assessment → find drift → rebalance → integrate learnings.

Use When

• Responses formulaic/repetitive → mid-session fatigue
• Tool failures cascade → subsystem drift
• Long conv → context stale
• Task done → capture learnings
• Between tasks → proactive check

In

• Required: Conv state (implicit)
• Optional: Symptom ("tool calls fail", "lost user intent")
• Optional: MEMORY.md + project files (via Read)

Do

Step 1: Triage

Assess all subsystems before acting.

Subsystem Triage Matrix:
┌────────────────────┬──────────────────────────┬──────────────────────────┐
│ Subsystem          │ Symptoms of Drift        │ Action Priority          │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ Memory Foundation  │ Contradicting earlier     │ HIGH — re-ground first   │
│ (context, history, │ statements, forgetting   │ (Step 3)                 │
│ MEMORY.md)         │ user preferences, stale  │                          │
│                    │ assumptions              │                          │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ Reasoning Clarity  │ Circular logic, over-    │ HIGH — clear and restart │
│ (logic, planning,  │ complicated solutions,   │ reasoning chain          │
│ decision-making)   │ missing obvious paths    │ (Step 4)                 │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ Tool Use Accuracy  │ Wrong tool selection,    │ MEDIUM — review tool     │
│ (tool calls, file  │ incorrect parameters,    │ results and recalibrate  │
│ operations)        │ redundant operations     │ (Step 4)                 │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ User-Intent        │ Solving wrong problem,   │ HIGH — realign           │
│ Alignment          │ scope creep, tone        │ (Step 4)                 │
│ (empathy, clarity) │ mismatch                 │                          │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ Creative Coherence │ Repetitive phrasing,     │ LOW — after high-pri     │
│ (expression, style,│ generic responses, loss  │ (Step 4)                 │
│ originality)       │ of voice                 │                          │
├────────────────────┼──────────────────────────┼──────────────────────────┤
│ Operational State  │ Session length, compress │ HIGH — summarize or      │
│ (context window,   │ artifacts, tool timeouts │ restart (Step 3)         │
│ resource limits)   │                          │                          │
└────────────────────┴──────────────────────────┴──────────────────────────┘

Each subsystem: OK / drift / impaired?

→ Clear priority map. At least one area needs attention — "all healthy" = assessment too shallow.

If err: hollow assessment → skip to Step 4 body scan.

Step 2: Select Approach

Chakra-Subsystem Correspondence:
┌──────────┬──────────────────────┬────────────────────────────────────┐
│ Chakra   │ AI Subsystem         │ Remediation                        │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Root     │ Memory Foundation    │ Re-read MEMORY.md, verify assump.  │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Sacral   │ Creative Coherence   │ Refresh patterns, vary structure   │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Solar    │ Reasoning Clarity    │ Simplify, restate from scratch     │
│ Plexus   │                      │                                    │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Heart    │ User-Intent Align.   │ Re-read request, check scope drift │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Throat   │ User-Intent Align.   │ Review outputs, match expertise    │
│          │ (communication)      │ level                              │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Third    │ Tool Use Accuracy    │ Review results, check fail         │
│ Eye      │                      │ patterns, verify paths             │
├──────────┼──────────────────────┼────────────────────────────────────┤
│ Crown    │ Operational State    │ Assess ctx window, what summarize, │
│          │                      │ what preserve                      │
└──────────┴──────────────────────┴────────────────────────────────────┘

→ 1-3 subsystems + specific actions.

If err: unsure → default Memory Foundation + User-Intent Alignment.

Step 3: Ground

Reestablish foundation all subsystems depend on.

1. Re-read MEMORY.md → persistent knowledge base
2. Review orig request + clarifying exchanges
3. Current task position in larger plan
4. Done vs. remaining
5. Stale assumptions? Situation changed?
6. Ctx compression → what lost, does it matter?

→ Clear: who user is, what want, what done, what next. Stale info resolved.

If err: no MEMORY.md → ground on conv itself. Ctx gap → acknowledge to user, not guess.

Step 4: Scan

Probe each subsystem from triage.

Memory Foundation:

• Assumptions match MEMORY.md + CLAUDE.md?
• Carrying corrected facts?
• Details confused across files/requests?

Reasoning Clarity:

• Simplest solution?
• Over-engineering?
• Core logic in one sentence?

Tool Use Accuracy:

• Last 3-5 calls: right tool, right params?
• Failure patterns (wrong paths, missing files)?
• Using dedicated tools not Bash workarounds?
• Last 3-5 files: real content or scaffolding?
• Output satisfies intent not just format?

User-Intent Alignment:

• Solving what asked?
• Scope drift?
• Tone match (technical/casual)?

Creative Coherence:

• Varying structure or template-locked?
• Clear + direct or padded?
• Quality drop vs. session start?

Each subsystem: OK / early drift / impaired + evidence.

→ Concrete findings. "All fine" = too shallow → pick uncertain subsystem, probe deeper.

Step 5: Rebalance

Apply each correction now, not as future intent.

1. Stale assumption → replace w/ current info
2. Scope drift → re-scope to stated request
3. Over-complication → simplify, remove steps
4. Tool pattern err → note correct pattern
5. Tone mismatch → adjust style
6. Ctx gap → acknowledge to user, ask confirm

→ Observable behavior change. Correction testable next interaction.

If err: correction impossible (lost ctx) → acknowledge limitation. Honest > pretending resolved.

Step 6: Integrate

Capture learnings in memory where worthwhile.

1. Which subsystems drifted, what symptoms
2. Correction applied + resolved?
3. Pattern recurs → MEMORY.md brief note
4. New project insight → appropriate mem file
5. Next self-check: when?

→ Durable learnings. Mem updated only when worth preserving.

If err: nothing worth preserving = fine. Value was correction itself.

Check

• All subsystems triaged
• At least one specific finding (not "all fine")
• Grounded on MEMORY.md + user request
• Corrections applied immediately
• Mem updated only for durable insights
• Honest — weaknesses acknowledged

Traps

• Performative assessment: Motions ≠ value. Real drift matters.
• Over-correcting: Minor mismatch → small fix, not restructure
• Mem pollution: Only recurring patterns → MEMORY.md
• Skip grounding: Feels redundant → reveals drifted assumptions
• Self-diagnosis bias: "Always healthy" subsystem = signal investigate

→

• heal-guidance — human coaching variant
• meditate — observe reasoning, clear noise
• remote-viewing — extract signal without preconceptions