安全
LYGO Context Guard — pre-flight token budget, secret redaction, and deterministic context compaction for AI agents. Use when stuffing tool dumps, logs, files, or long chats into a model; when context is too large; when you need to save tokens / reduce API cost; when you must redact API keys before re-injection. Pure local stdlib. No network, no subprocess. Commands: estimate, redact, compact, budget, toolpack, preflight. Install clawhub:@deepseekoracle/lygo-context-guard.
LYGO Context Guard — pre-flight token budget, secret redaction, and deterministic context compaction for AI agents. Use when stuffing tool dumps, logs, files, or long chats into a model; when context is too large; when you need to save tokens / reduce API cost; when you must redact API keys before re-injection. Pure local stdlib. No network, no subprocess. Commands: estimate, redact, compact, budget, toolpack, preflight. Install clawhub:@deepseekoracle/lygo-context-guard.
The pre-flight every agent needs.
OpenClaw burns money when tool dumps and chat history explode the context window.
Context Guard is a local lattice gate: estimate tokens → redact secrets → compact deterministically → check budget — before you call the model.
Signature: Delta9Phi963-CONTEXT-GUARD-v1.0.0
ClawHub: @deepseekoracle/lygo-context-guard
| Problem | What agents do wrong | Guard does |
|---|---|---|
| Token burn | Re-inject full logs/tool JSON every turn | Compact + budget exit code |
| Secret leaks | API keys in tool output → model context | Redact patterns |
| Context overflow | “Just paste the whole file” | Head/tail + mid map + SHA |
| No visibility | Guessing token counts | Dual heuristic estimate |
Searches this skill targets: token saver, context window, compress context, reduce tokens, prompt budget, redact secrets, tool result too large.
npx clawhub@latest install deepseekoracle/lygo-context-guard
cd path/to/lygo-context-guard
python scripts/self_check.py
# Estimate tokens (heuristic)
python scripts/context_guard.py estimate --text "long tool dump..."
python scripts/context_guard.py estimate --file ./huge_log.txt
# Redact secrets (keys, JWTs, private key blocks, password=...)
python scripts/context_guard.py redact --file ./tool_out.txt
# Deterministic compact (dedupe lines + head/tail keep)
python scripts/context_guard.py compact --file ./tool_out.txt --max-chars 8000
# Budget gate (exit 10 if over)
python scripts/context_guard.py budget --file ./pack.txt --budget 4000
# One-shot for agents: redact + compact + budget
python scripts/context_guard.py toolpack --file ./tool_out.txt --budget 4000 --max-chars 8000
# Full preflight report
python scripts/context_guard.py preflight --file ./blob.txt --budget 8000
python scripts/context_guard.py preflight --file ./blob.txt --write last_preflight.json --i-consent
# Demo (synthetic leaky dump)
python scripts/context_guard.py demo
| Command | Network | Writes | Notes |
|---|---|---|---|
estimate / redact / compact / budget / toolpack | none | no | stdout JSON + text |
preflight --write | none | skill state/ only | needs --i-consent |
demo | none | no | built-in sample |
Before re-injecting a large tool result or log:
1. Save tool output to a temp file
2. python scripts/context_guard.py toolpack --file THAT_FILE --budget 4000
3. If exit code 10 → lower max-chars or split the task
4. Feed packed_text back into the model — not the raw dump
Exit codes:
| Code | Meaning |
|---|---|
| 0 | OK / under budget |
| 10 | Over budget after pack |
| 2 | Missing file / bad input |
| Skill | Role |
|---|---|
lygo-api-token-saver | Broader pay-to-go budget ops + local army preference |
lygo-pxpipe-lygo | Vision / multi-tool dump compression |
lygo-ops-detector | Discourse signal analysis (different problem) |
lygo-kickstart-wizard | Onboarding map |
See references/SECURITY.md.
Δ9Φ963 — compress chaos · redact secrets · budget the lattice · human remains the publisher.
LYGO CLI Bridge — unified entrypoint: lygo health | map | analyze | mint | radar | next. Wraps kickstart + ops-detector + mint-walkthrough + deception-radar without exposing internal skill layout. In-process imports only; no subprocess; no auto-publish. Install clawhub:@deepseekoracle/lygo-cli-bridge.
Multi-tool vision context compression for LYGO (Anthropic, OpenAI, Grok/xAI, Gemini). Use before stuffing huge prompts or logs into pay-to-go APIs. Requires...
LYGO Continuum — falsifiable work capsules for AI agents and humans. Seal 'done' as checkable claims (file SHA-256, contains, JSON paths, globs under --base), re-verify, drift, handoff. CLI: pure local stdlib — no network, no subprocess. Paths/globs confined under --base; --out under base or state/ with --i-consent. Optional separate browser portal is NOT opened by the skill. Commands: seal, verify, drift, handoff, card, demo. Install clawhub:@deepseekoracle/lygo-continuum.
Minimize pay-to-go xAI/API token spend. Use when user says token saver, save tokens, API budget, pay-to-go burn, or before long Grok/Claude/GPT calls. Prefer...
LYGO Deception Radar — public proof layer for Ops Detector. Rebuilds an anonymized radar feed + HTML dashboard from the public labeled discourse suite only. Shows strong vs weak vs clear bands with operational threshold 0.65. Not for doxing or private mail. No subprocess. Install clawhub:@deepseekoracle/lygo-deception-radar.
OpenClaw plugin skill — live Haven pulse, stack verify, registry compare, star chart gate, alignment readiness. Install clawhub:@deepseekoracle/lygo-lattice-pulse.