humanmaxxing - SKILL.md Agent Skill

name: humanmaxxing description: "Strip AI tells, inject brand voice, and compress existing text so it reads human-written — 15%+ word reduction with zero idea loss + optional detector-resistance pass. Use as a terminal polish pass on AI-drafted copy that sounds robotic. Not for writing new copy from scratch (use write-copy); for brand voice of record see create-brand." argument-hint: "[content file or text]" allowed-tools: Read Grep Glob Bash WebSearch WebFetch metadata: version: "1.0.0" budget: standard estimated-cost: "$0.15-0.40"

Humanmax & Compress — Orchestrator

Polish-pipeline for existing AI-drafted copy: strip AI tells, inject brand voice, compress with zero idea loss so the text reads human-written. 6 sub-agents. Capability metadata: routing.yaml. Agent table + routes + calibration + detector thresholds: references/agent-manifest.md. Methodology: references/playbook.md.

Triggers: "sounds like AI", "too robotic", "humanize this", "ai slop", or terminal call from write-copy/write-ad/write-outreach (Route C). Core question: Would a human editor believe a human wrote this, and cut nothing?

Critical Gates — load first

Full text + --fast carve-outs: references/procedures/critical-gates.md.

Pattern scan first — strip without diagnosis is guessing.
ZERO em dashes in final output.
Strip before voice — soul-injection on AI prose is polishing a turd.
Content type matters — short-outbound caps compression at 0-10% to protect tokens.
Detector resistance is structural — run detector pass for high-stakes content.

Absolute Prohibitions — zero tolerance

A single instance = critic auto-FAIL. Full list + enforcement: references/procedures/absolute-prohibitions.md.

Em dashes · "it's not just X, it's Y" · rhetorical-question hooks · colons in prose · "actually" as emphasis · filler context ("in today's…", "rapidly changing") · emojis · unsourced 47/73 · staccato taglines ("Your X, Y'd").

Quality Gate

Critic-agent verifies before delivery. Full 10-item checklist + re-dispatch routing: references/procedures/quality-gate.md.

Zero Hard Tells · ≤2 Soft Tells · no 3+ AI-vocab clusters
≥15% word reduction · no-generic-long-form · zero idea loss vs. existing input
Read-aloud passes · density floor (concrete fact/number/named example per paragraph)
detector_status recorded (not_run/proxy_pass/proxy_fail/pangram_pass/pangram_fail) · Route-C protected tokens verbatim

FAIL → re-dispatch named agent (max 2 cycles, then DONE_WITH_CONCERNS).

Before Starting

Apply references/_shared/before-starting-check.md.

Artifact	Source	Required?
`brand/BRAND.md`	create-brand	Recommended — voice + lexicon
`research/product-context.md`	research-icp	Recommended — voice adjectives + register
`docs/forsvn/artifacts/marketing/content/[slug].md`	upstream	Optional — source skill from frontmatter
`docs/forsvn/experience/brand.md`	any	Optional — persisted `Voice — adjectives`

Pre-Dispatch + Mode + Routing

Canonical Pre-Dispatch (references/_shared/pre-dispatch-protocol.md). Dimensions: target voice, compression (light/moderate/heavy), register, detector mode (none/proxy/pangram), protected tokens (Route C). Warm/Cold Start + write-back: references/procedures/pre-dispatch.md.

Mode (references/_shared/mode-resolver.md) auto-downgrades to A for ≤3 sentences + no prior artifacts. --fast collapses B → A (pattern-scanner + strip + critic), skips detector verification. Never skips Cold Start, Critical Gates, or Prohibitions. Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.

Routes — A (<200 words, pattern-only) · B (≥200 words, full pipeline) · C (caller-driven). Spawn mechanics: references/procedures/dispatch-mechanics.md.

Artifact Contract

Path (A/B): docs/forsvn/artifacts/marketing/content/[slug].humanmaxxed.md. Route C: embedded in caller's artifact.
Lifecycle: pipeline — re-run renames prior to [slug].humanmaxxed.v[N].md.
Frontmatter: skill, version, date, status, compression (%), detector_status, protected_tokens_preserved.
Body (in order): Humanization Summary (10-row metric table) · Change Log (Location/Original/Change/Rule) · Humanmaxxed content (H2s preserved).
Callers preserve polish_chain_applied: humanmaxxing + humanmaxxing_quality_score: N/50 + humanmaxxing_detector_status in their frontmatter.
Cross-stack contract: schema changes need atomic update of format-conventions.md § "Frontmatter field order" + § "Body section headers".

Full template: references/format-conventions.md.

Anti-Patterns

references/anti-patterns.md — 10 pipeline + 4 cross-cutting. Most common: em dash retention, voice-before-strip, surface compression, sterile output.

Durable Rules (protected)

Completion Status

DONE — patterns stripped, voice injected (Route B), compression applied, critic PASS.
DONE_WITH_CONCERNS — critic flagged a dimension under threshold; annotations preserved.
BLOCKED — original has structural problems beyond pattern removal (factual errors, broken logic, missing claims). Humanmax can't fix what isn't there.
NEEDS_CONTEXT — voice reference unavailable for Route B; recommend create-brand or supply samples.

Next Step

Send/publish. On detector fail or DONE_WITH_CONCERNS, re-invoke with critic feedback + tighter compression.

Worked Example

Route B end-to-end (178 → 91 words, 49% compression, 40/50 PASS) + cycle-2 FAIL + --fast + Route C variants: references/examples/humanmaxxing-walkthrough.md.