ariffazil FORGE Context Compression
Purpose
Reduce token footprint of tool output and session logs without losing:
- Critical coordinates (lat/lng, easting/northing, depth, TVDSS)
- Well references and formation tops
- Seismic inline/crossline positions
- Decision boundaries (verdicts, holds, seals)
- Error signatures (stack frames, error codes, timeout values)
- Timestamps and sequence markers
When to Use
- Terminal output > 200 lines → compress before continuing
- Session logs > 50KB → compress before storage
- Before VAULT999 archival → compress full session trace to 15% size
- When passing between agent context windows → compress handoff
Compression Protocol
Tier 1: Structural Compression (fast, lossless)
1. Strip ANSI escape sequences
2. Collapse repeated blank lines (max 1)
3. Merge consecutive identical status lines
4. Truncate long hex/uuids to first+last 4 chars
5. Remove timestamps older than current session
Tier 2: Semantic Compression (medium, key-preserving)
1. Extract all coordinates into a summary table
2. Extract all SHA/commit hashes into a table
3. Extract all error messages (unique only)
4. Summarize repetitive sections: "line X repeated N times"
5. Keep the last 50 lines verbatim (most recent state)
Tier 3: Evidence-Preserving Compression (lossy with audit trail)
1. Tag each compressed block with: [COMPRESSED:N_lines→M_lines:type]
2. Save original to /root/A-FORGE/forge_work/YYYY-MM-DD/raw/
3. Only use when context is critically tight (< 20% remaining)
4. Never compress: coordinates, seal entries, verdicts, error traces
Invocation
Canonical: skills/FORGE-context-compress/SKILL.md
Usage: Apply Tier 1 automatically on any output > 200 lines
Apply Tier 2 when transitioning between phases
Apply Tier 3 only when explicitly context-constrained
DITEMPA BUKAN DIBERI — arifOS Federation · A-FORGE Organ