name: memory-config description: View and modify ConKeeper memory configuration settings. Use to adjust token budget, output style, and other preferences after memory initialization.

Memory Configuration

View and modify ConKeeper configuration for the current project.

Pre-flight Check

1. Verify .claude/memory/ exists
   • If not: Inform user to run /memory-init first

Configuration Options

Token Budget Presets

Other Settings

Workflow

Step 1: Read Current Config

Check for .claude/memory/.memory-config.md:

• If exists: Read and display current settings
• If not: Display defaults (standard preset)

Step 2: Display Current Settings

Current ConKeeper Configuration

• Token budget: [economy/light/standard/detailed] (default: standard)
• Suggest memories: [true/false] (default: true)
• Auto load: [true/false] (default: true)
• Output style: [quiet/normal/explanatory] (default: normal)
• Auto-sync threshold: [0-100] (default: 60)
• Hard-block threshold: [0-100] (default: 80)
• Context window tokens: [integer] (default: auto-detected from model, fallback: 200000)
• Observation hook: [true/false] (default: true)
• Observation detail: [full/stubs_only/off] (default: full)
• Correction sensitivity: [low/medium] (default: low)
• Auto-reflect: [true/false] (default: true)
• Staleness commits: [integer] (default: 5, 0 = disable)
• Project search paths: [absent/disabled/paths] (default: absent)

Step 3: Ask What to Change

Ask the user which setting they'd like to change, or whether they're done. Accept natural language responses (e.g., "change output style to quiet", "disable auto-reflect").

Step 4: Apply Changes

Update or create .claude/memory/.memory-config.md:

---
token_budget: standard
suggest_memories: true
auto_load: true
output_style: normal
auto_sync_threshold: 60
hard_block_threshold: 80
context_window_tokens: 200000
observation_hook: true
observation_detail: full
correction_sensitivity: low
auto_reflect: true
staleness_commits: 5
# project_search_paths: ["~/zed", "~/work"]  # absent = off, disabled = permanently off
---

Observation Hook Settings

• full: Full entries for Bash/external tools, stub entries for native tools
• stubs_only: Stub entries for all tools (timestamp, tool, type, path, status only)
• off: No observation logging (same as observation_hook: false)

Auto-detection: If context_window_tokens is not explicitly set, ConKeeper reads ~/.claude/settings.json to detect the active model's context window. Models with the [1m] variant (e.g., opus[1m]) use a 1,000,000 token window. All others default to 200,000. Set context_window_tokens explicitly to override auto-detection.

Correction Detection Settings

• low: Conservative patterns only (fewer false positives, higher precision)
• medium: Adds looser patterns like "instead", "should be", "rather"

Note: high sensitivity was intentionally omitted — Claude Code's facets data provides higher-accuracy retrospective friction classification. This hook is a fast first-pass; /memory-reflect uses facets for accurate second-pass analysis.

Create .correction-ignore in project root to suppress specific patterns:

# Patterns to never flag as corrections
# One line per literal substring, matched case-insensitively
no worries
try again with verbose

Reflection Settings

Session depth (LIGHTWEIGHT vs STANDARD) is auto-detected based on observation and correction counts.

Privacy Tags

Privacy tags are always enforced — there is no configuration toggle.

• Wrap sensitive content in <private>...</private> tags
• Add private: true to YAML front matter for entire-file privacy
• Private content is excluded from context injection, search, sync, and reflection

Step 5: Confirm

Configuration updated.

• [Setting]: [old value] → [new value]

Changes take effect in the next session.