Paperthin - Setup Guide¶

Source: skills.sh (22.1K combined installs) GitHub: lilmgenius/paperthin (945⭐, MIT) Category: Agent Operating Patterns & Context Discipline First Seen: June 21, 2026 (skills.sh) Quality Tier: 🟢 Production (Gen Agent Trust Hub Pass / Socket Pass / Snyk Pass)

"Low-level agentic design patterns. Turning old engineering wisdom into reflexes your agent reaches for on its own." Paperthin is a discipline toolkit for long-running agent work: ssotize enforces single-source-of-truth with a read-only audit → approval → consolidation gate; debloat trims context bloat; factchk and readchk verify claims and reading comprehension; the re0 family ("ground zero") resets, re-plans, and re-targets work when a session drifts. Every skill is small, approval-gated where it mutates artifacts, and designed to become reflex.

Installation¶

npx skills add lilmgenius/paperthin --skill ssotize    # SSOT consolidation
npx skills add lilmgenius/paperthin --skill debloat    # context/artifact bloat removal
npx skills add lilmgenius/paperthin --skill factchk    # claim verification
npx skills add lilmgenius/paperthin --skill feynman    # plain-language explanation
npx skills add lilmgenius/paperthin --skill re0        # ground-zero reset workflow

Core Skills¶

Skill	Installs	Use For
ssotize	858	Single-source-of-truth consolidation (audit → approve → consolidate)
shower / hate / sip	855 / 844 / 829	Unblocking and judgment resets for stuck sessions
re0 + re0-git/work/release/memo/plan/loop/upgrade/merge	708-851	Ground-zero family: reset, re-plan, re-target work mid-task
factchk	842	Fact-checking claims before they enter output
mandela	841	Detecting false-memory / assumed-fact drift
readchk	740	Reading-comprehension verification of source material
modelchk	725	Model capability/limitation self-check
feynman	724	Feynman-technique plain-language explanations
prism	722	Multi-perspective analysis
debloat	719	Context and artifact diet
catchup / macrothink / aim	740 / 728 / 712	Session recovery and goal alignment
dedash / detool / reorder	821 / 717 / 713	Refactoring agent habits and workflows
ssotchk / retro / flywheel	122 / 111 / 110	SSOT health, retrospectives, compounding improvement
redteam / scratch / tasting	5 / 3 / 10	Adversarial and experimental edge workflows

Prerequisites¶

None - pure prompt-skill toolkit, no API keys or services
Works with any agent runtime that loads skills.sh skills

CorpusIQ Use Cases¶

SSOT discipline - ssotize's audit-then-mutate gate is the exact pattern CorpusIQ uses for brand facts and product claims across content surfaces
Context token economics - debloat and readchk reinforce the field-filtering and memory-compaction doctrine already in CorpusIQ-connected agent operating rules
Research rigor - factchk/mandela/prism harden competitive-intelligence sweeps against hallucinated or assumed facts
Long-session recovery - the re0 family is a reusable playbook for resetting stalled multi-hour agent sessions

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations / Verification¶

Pattern skills encode discipline, not tooling - value depends on the agent actually invoking them mid-task
Some skill names are opaque until read; the SKILL.md files are short and worth a one-time skim
Verify install: npx skills list | grep -E "ssotize|debloat|factchk"

Security¶

Gen Agent Trust Hub: Pass · Socket: Pass · Snyk: Pass