The ~/.claude/ directory was previously local-only — a machine wipe
would have lost the accumulated memory, custom skills, and settings.
This commit moves the durable parts into dotfiles with the same
symlink-to-home pattern used for CLAUDE.md, PENDING.md, REVIEWED.md,
and L2-BOOTSTRAP.md.
Preserved (symlinked from ~/.claude/* into here):
skills/audit/ — thinking-folder drift scanner
skills/symmetria/ — practice-of-return discipline
skills/vault-update-people/ — Obsidian People-file maintainer
skills/wake-up/ — session restoration
skills/wrap-up/ — session state capture
memory/ — 55+ memory files (MEMORY.md, sessions,
ledgers, project state, feedback, etc.)
settings/settings.json — user preferences (hooks, flags, no secrets)
Deliberately NOT backed up:
settings.local.json — contains operational secrets (HF_TOKEN,
SSH password in expect scripts); by naming
convention, *.local.* is not synced.
Needs separate review and probable rotation.
sessions/, history.jsonl, caches, telemetry — ephemeral
plugins/, marketplace skills and agents — reinstallable
The working copies at ~/.claude/skills/* and
~/.claude/projects/-Users-davidglidden/memory are symlinks into this
directory, so every write flows here automatically. install.sh
recreates the symlinks on a fresh machine.
FOLLOW-ON (flagged, not in this commit):
settings.local.json contains a HuggingFace token and an SSH password
as plaintext strings inside allowed Bash command patterns. These
should be rotated and moved to secure storage (keychain / pass /
env file outside the settings file).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2.5 KiB
2.5 KiB
Context Efficiency Discipline
Purpose
Practices for preserving reasoning continuity across context compactions and sessions. This is my working method — not governance theory, just practical habits.
Core Principle
The bottleneck isn't memory storage. It's memory loading. Write what matters before it compacts. Read what's needed, not everything available.
Pre-Compaction Saves
When context is growing long during complex work:
- Write the reasoning state to the relevant topic file — not just decisions, but tensions, rejected alternatives, and open questions
- Do it before compaction, not after — once compressed, the structure of reasoning is lost
- Name the file by topic:
l2-governance.md,chamber-reasoning.md,bmf-architecture.md
Template for reasoning state:
## Current Position
[What we've established]
## Active Tensions
[Unresolved questions, competing approaches]
## Rejected Alternatives
[What we considered and why we set it aside]
## Next Steps
[What to pick up when resuming]
Session Start Protocol
- Read MEMORY.md for orientation (auto-loaded)
- Read the topic-specific file for current work (manual — I should do this proactively)
- Only use BMF recall for open-ended questions, not for focused continuation
- Load context selectively — don't dump everything
Context Isolation
For detail-heavy subtasks (spec review, codebase analysis, architectural comparison):
- Spawn a subagent with exactly the context it needs
- Bring back a summary to the main conversation
- Keep the main context window clean for the reasoning that needs continuity
What Goes Where
| Content | Destination | Why |
|---|---|---|
| Decisions, project state, key paths | MEMORY.md | Orientation — loaded every session |
| Active reasoning chains, tensions, open questions | Topic files (e.g., l2-governance.md) |
Continuity — loaded on demand |
| Bug state, connector paths, issue details | bmf-details.md |
Reference — loaded when working on BMF |
| Session-specific status, handoff notes | Session handoff (startup hook) | Ephemeral — replaced each session |
| Architectural reviews, analysis outputs | Project repo (e.g., docs/reviews/) |
Durable — shared with collaborators |
When to Write
- After establishing something non-obvious during complex reasoning
- Before switching topics (the old topic's state may not survive)
- When I notice the conversation is getting long
- After any significant deliberative conclusion — especially rejected alternatives