Lossless Compaction Protocol

Lossless Compaction Protocol ("prepare for compact")

In extended autonomous coding shifts spanning dozens of turns, AI context windows inevitably fill up. Environments like Claude Code (/compact), Cursor resets, or Gemini CLI context rollovers periodically summarize the chat log.

When an un-scaffolded agent undergoes compaction, it suffers from Compaction Amnesia: active hypothesis chains, test states, and mental cartography are wiped out. The agent wakes up on post-compact Turn 1 confused, prone to regressions, and repeating measured errors.

SDCS eliminates Compaction Amnesia through the Mid-Shift Checkpoint Protocol ("prepare for compact") and the Wiring Mutation Invariant (Pillar 2).

---

The 4 Operational Dimensions

Dimension	Specification Contract	Operational Behavior
WHY	Lossless Memory Persistence	Guarantees zero context loss. Fine-grained history is externalized to the immutable flight recorder, while active working memory is trimmed to $\le 300$ tokens so Turn 1 hydration is immediate.
WHEN	1. Human Trigger: "prepare for compact" 2. Token Saturation (70–80%) 3. Milestone Completion	Triggered mid-shift whenever the developer types "prepare for compact" before running /compact, or when token usage nears context exhaustion.
WHERE	Cross-Pillar Synchronization: • wiring.yaml (Pillar 2) • sessions/*.md (+1 Flight Recorder) • state.md (Pillar 4) • decisions.md (Pillar 6) • app_map.md (Pillar 5) • roadmap.md (Pillar 3)	Checkpoints are written to physical disk files before memory is cleared.
HOW	5-Step Mechanical Checklist	The agent mechanically executes the 5-step checklist and emits an explicit readiness signal before compaction proceeds.

---

The 5-Step Checkpoint Execution Checklist

[Developer: "prepare for compact" or Token Saturation Nears]
                             │
                             ▼
  1. Topology Audit              ──► Runs `sdcs verify --topology` to audit wiring.yaml
                             │
                             ▼
  2. Flight Recorder Snapshot    ──► Writes immutable log to `sessions/YYYY-MM-DD_<topic>.md`
                             │
                             ▼
  3. Blackboard Pruning          ──► Prunes `state.md` strictly to ≤ 300 tokens
                             │
                             ▼
  4. Episodic Memory Sweeps      ──► Records failed approaches to `decisions.md` & runs `sdcs map --sync`
                             │
                             ▼
  5. Compact Readiness Signal    ──► Emits confirmation: "Ready for compaction."

1. Topology & Subsystem Audit

Run sdcs verify --topology to verify that all imports comply with wiring.yaml. If new modules or packages were created during the shift, ensure they are declared in wiring.yaml.

2. Flight Recorder Checkpoint

Write an immutable checkpoint log to sessions/YYYY-MM-DD_<topic>.md capturing:

• Work completed during the shift.
• Current verification status and gate standing.
• Active blockers or open questions.
• Immediate post-compact next steps.
• Update sessions/manifest.jsonl via sdcs session index.

3. Blackboard Pruning (state.md)

Aggressively prune and overwrite state.md strictly to $\le 300$ tokens containing only:

## Current Objective
[Concise active milestone goal]

## Status & Gate Verification
[Passing gates: Gate T, Gate M, Gate C, Gate A, Gate E]

## Immediate Next Action (Post-Compact)
[Next exact file to edit or test to run]

4. Episodic Memory Sweeps

• Log any rejected approaches or failed experiments to decisions.md.
• Synchronize app_map.md if files were created, moved, or deleted (sdcs map --sync).
• Update roadmap.md [MEASURED] blocks if milestones or acceptance criteria were met.

5. Readiness Signal

Output a brief confirmation:

"All 7 pillars and the flight recorder are synchronized. Ready for compaction."

The developer then triggers /compact or restarts the session. On post-compact Turn 1, the agent reads the clean $\le 300$-token state.md and resumes engineering with zero amnesia.