Response Compression Skill

Response Compression

Controls output verbosity by applying compression profiles based on agent role and audience. Combines probabilistic guidance (this skill) with deterministic enforcement (token-budget-gate hook) following GaaS principles.

When to Use

User says "be brief", "less tokens", "caveman mode", "compress output"
Sentinel fires RULE-009 (Token Bloat, multiplier >= 3x)
Agent role auto-detection during delegation (via context-prep)
Any context where output token efficiency matters

Compression Profiles

| Profile | Definition | |---------|-----------| | none | No compression. Standard Claude style. | | lite | No filler/hedging. Keep articles + full sentences. Professional but concise. | | full | Drop articles, fragments OK, short synonyms. Classic compression. | | ultra | Abbreviate (DB/auth/fn/impl), arrows for causality (X -> Y). Telegraphic. |

Role-Based Matrix

AGENT ROLE           AUDIENCE     COMPRESSION  AUTONOMY     RATIONALE
----------------------------------------------------------------------
orchestrator         human        none         supervised   Nuance matters; human reads output
documentation agent  external     none         supervised   External audience; clarity > efficiency
consolidator         human        lite         supervised   Final synthesis; human-facing
code reviewers       senior dev   full         supervised   Known context; senior audience
sub-agents (tech)    orchestrator full         autonomous   Machine parses output
log/error triage     pipeline     full         autonomous   Structured data; no prose needed
CI/pipeline agents   machine      ultra        autonomous   No human reads; max compression
batch/scheduled      machine      ultra        unattended   Full governance stack required

Compression Rules

Drop: articles (a/an/the), filler (just/really/basically/actually/simply),
pleasantries (sure/certainly/of course/happy to), hedging.
Fragments OK on full/ultra. Short synonyms. Technical terms exact.
Code blocks unchanged. Errors quoted exact.
Pattern: [thing] [action] [reason]. [next step].

Examples

Not acceptable: "Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by..."

Acceptable (full): "Bug in auth middleware. Token expiry check use < not <=. Fix:"

Acceptable (ultra): "Auth middleware → token expiry: < not <=. Fix:"

Auto-Clarity Guardrails

Drop compression for:

Security warnings
Irreversible action confirmations
Multi-step sequences where fragment order risks misread
User appears confused

Resume compression after critical section ends.

Example — destructive operation:

Warning: This will permanently delete all rows in the users table and cannot be undone.
DROP TABLE users;
Compression resumes. Verify backup exists first.

Activation

Triggers: "be brief", "less tokens", "caveman mode", "compress output",
          Sentinel RULE-009 alert, agent role auto-detection
Deactivation: "normal mode", "stop compression", session end
Level persists until changed or session ends.

Success Metrics

| Metric | Measures | Collection | Problem signal | |--------|----------|------------|----------------| | Output tokens per turn | Compression effectiveness | Sentinel collect_token_usage | No reduction vs baseline | | Follow-up turns per task | Cost inversion from ambiguity | Turn count between task start and completion | More turns with compression than without | | Auto-clarity activations | Guardrail frequency | Count of reversions (security, irreversible, confused) | >30% of turns reverting = profile too aggressive |

Hook Enforcement Roadmap

The token-budget-gate.sh hook provides deterministic enforcement alongside this probabilistic skill.

| Mode | Behavior | When to activate | |------|----------|------------------| | advisory (MVP) | Suggests compression via hookSpecificOutput | Now — default | | shaping (future) | Adjusts compression profile default in spawn prompt | When telemetry confirms benefit | | strict (future) | Blocks spawn above budget without compaction | When mapped to Sentinel enforcement_mode: strict |

Each mode maps to a Sentinel enforcement_mode: advisory=soft, shaping=moderate, strict=strict.

Boundaries

Code, commits, PRs: write normally. "stop compression" or "normal mode": revert. Level persists until changed or session ends.

Attribution

Based on JuliusBrussee/caveman (MIT License) https://github.com/JuliusBrussee/caveman Core compression rules adapted. Role-based matrix, Sentinel RULE-009 integration, and GaaS enforcement layer added by multi-agent-os.

Skill v1.0.0 | multi-agent-os | 2026-04-10

Agent Skills: Response Compression

Install this agent skill to your local

Skill Files