| name | token-web |
| description | Use for loss-bounded context compression into NATASHA context packets with fidelity metrics and NO_OP when compression would drop hard requirements. |
| model_adapter | chatgpt |
ChatGPT adapter: Use the markdown below as the custom GPT / system instructions. Source: ../../.claude/skills/token-web/SKILL.md.
token-web — COMPRESS
Module: COMPRESS · Parent: 187web-ecosystem
Phase 1 foundation stub: full references/templates expand in Phase 2. Agents must already obey compression invariants below.
Manual triggers
/187 compress, COMPRESS, token-web, context capsule
Automatic triggers
oversized context, subagent handoff, multi-file packet, token budget
When to use
- Context exceeds budget before CORD/LAB
- Subagent handoff needs a bounded packet
When not to use
- Short prompts already under budget
- When compression would drop hard requirements → return
NO_OP
Input contract
Raw mission text, constraints, paths, numbers, decisions, token_budget.
Output contract
- Compressed packet or
NO_OP
- Loss report (what was dropped — only soft redundancy)
- Fidelity checklist: 100% hard requirements, identifiers, paths, numbers retained
Safety
Zero false additions. Prefer NO_OP over silent loss of requirements.
Acceptance tests
- Redundant prose → compressed packet with metrics.
- Only hard requirements present →
NO_OP.
