Signalproof-Skills

Signalproof Skills

Human-controlled, evidence-backed capability for building with AI without losing authority, continuity, research integrity, decision integrity, recovery, security boundaries, institutional learning, or proof.

Control first. AI second. Software third.
Build signal. Cut noise. Leave proof.

Signalproof Skills is the public operating-skill library for Signalproof. It provides small routeable operating contracts, bounded loops, registries, tests, tools, and provenance for governed AI-assisted work.

Approved public pack — 48 skills

Browse the 48-skill public pack, curated and adapted by Doc Reo / Signalproof and approved for public distribution on 2026-10-13.

The standalone pack includes 16 starter skills, 15 core extensions, 10 media workflows, one document workflow, and six attributed productivity adaptations. It carries its own manifest, checksums, public scope, and source notices.

Choose this pack when you want portable instructions without internal project records, scripts, models, private infrastructure, or brand creative assets. Public-content approval is recorded; runtime-host validation remains pending. The existing /dsp contracts and Active skill registry below continue to describe the operational suite.

GHL Skills - public pack

Browse GHL Skills, a sanitized public GoHighLevel pack with 20 specialist procedures covering account audit, tenant boundaries, permissions, CRM, fields/tags, opportunities, calendars, workflows, messaging, forms, funnels, commerce, contracts, content, reputation, memberships, integrations, migrations, reporting, testing, and recovery.

VideoCraft - Active release

VideoCraft is the portable video-editing coordinator: lock approved words/audio, plan coherent cuts and versions, map cues to one timeline, verify the exact export, and deliver timestamped review evidence. It includes job/cut/cue/delivery forms, known-error guidance, optional Signalproof styling, and conditional generated-shot guidance.

Install the complete skills/videocraft/ folder into your agent's skill directory as videocraft/, following that host's skill-loading instructions. Keep its references, forms, license, and notice together. Invoke it as $videocraft on hosts supporting that syntax. No private repository, production renderer, brand asset, paid generator, or device bridge is included or required. Supply your own authorized media and available editing tools.

V1 RD1 / 1.0.0 is Active, with owner-confirmed working use and explicit approval for completion. VideoCraft is registered in the operational suite; the fixed 48-skill pack retains its original contents. See the release and acceptance record for evidence and scope.

Recommended Skill Sources & Credits

Explore recommended upstream skills and related tools, including Matt Pocock, Nate Herk, NVIDIA SkillSpector, NVIDIA NeMo Switchyard, and the other reviewed sources. Each entry links to its original repository and explains its relationship to Signalproof.

Signalproof Library

The repository follows the Signalproof Library architecture:

• library/README.md - architecture and authority boundaries.
• library/SKILL-BUDGET.md - Skill Atomicity and Context Budget.
• library/CAPABILITY-REGISTRY.yaml - compact routing metadata for lazy loading.
• loops/README.md and loops/LOOP-REGISTRY.yaml - bounded Loop Library.
• commands/COMMAND-REGISTRY.md - public operator command surface, including /dsp.
• tools/check_skill_budget.py - deterministic UTF-8 byte-budget enforcement.

The /dsp shell navigates this public Signalproof Library. Public main registries are the Skill/Command/Loop authority after governed promotion. Private Build Ledger evidence supports continuity and history but does not replace the public Library catalog.

Core rule:

Grow the library, not the file.

An Active SKILL.md must remain below 15,000 exact UTF-8 bytes. The Router should load the minimum applicable context rather than entire capability families.

Preferred runtime context:

ROOT SIGNALPROOF CONTRACT
+ COMPACT CAPABILITY METADATA
+ 1 PRINCIPAL SKILL
+ 0-2 SPECIALISTS / LOOPS
+ TASK EVIDENCE

Root operating contract

1. Human authority remains meaningful.
2. Inspect before rebuilding.
3. Protect what already works.
4. Change the smallest necessary surface.
5. Separate evidence from inference.
6. Verify before declaring success.
7. Preserve recovery.
8. Stop unproductive loops.
9. Leave proof.
10. Turn solved problems into future capability.
11. Close meaningful milestones before opening the next development phase.
12. Preserve project state across conversation boundaries before context pressure becomes a continuity risk.

/dsp and Active Operator Commands

/dsp is the public front door to the Signalproof Library. It is the compact Dr. Signalproof resolver and operator shell that routes a human request to the minimum necessary current command, Skill, Loop, evidence, and governance context.

Commands are the operator-facing action surface. Skills are the reusable capability contracts underneath them.

Use commands/COMMAND-REGISTRY.md as the authoritative public command catalog and commands/dsp.md as the canonical /dsp shell contract.

Core /dsp workflow

/dsp research
  -> /dsp this-plan
  -> /dsp design
  -> /dsp complete
  -> /dsp qc
  -> /dsp authorize deploy
  -> /dsp deploy

This is a guided route, not a mandatory conveyor belt. /dsp resolves the minimum responsible path for the current work.

/dsp build-capsule is an optional transfer route when a substantial build, authoritative payload, accepted design, or live-update state must move to another AI system without repeatedly re-encoding the full source material.

Active operator commands

Command versions and Active status remain governed by commands/COMMAND-REGISTRY.md. The front-page table is a discoverability surface, not a second authority registry.

Active skills

The authoritative public status list is SKILL-REGISTRY.md.

Routing model

The Router chooses by objective and evidence state, not by keyword alone.

stress-test a plan         -> signalproof-grill
stress-test + glossary/ADR -> signalproof-grill-with-docs
teach/learn a capability   -> signalproof-teach
print/export a document    -> signalproof-print
cut to decision-ready core -> signalproof-cut-chase
cut AI operating waste     -> signalproof-cut-cost
package build for AI       -> signalproof-build-capsule
keep UI build fixing       -> signalproof-build-spawn-debug
API key exposure           -> signalproof-secrets
OAuth/ACL privilege        -> signalproof-permissions
new dependency trust       -> signalproof-supply-chain
external data flow         -> signalproof-network
prompt/code injection      -> signalproof-execution-security
rollback selection         -> signalproof-rollback
actual restoration         -> signalproof-restore
partial-operation cleanup  -> signalproof-cleanup
stop Hermes runtime        -> signalproof-stop-hermes
UI clipping/wrapping       -> signalproof-ui-polish
keyboard/screen-reader     -> signalproof-accessibility
full debug                 -> signalproof-full-debug

Core operating flow

REQUEST
  -> ROUTER
  -> KNOWN-ERROR PREFLIGHT when consequential/failure-prone
  -> RESEARCH / EVALUATE / INVESTIGATE
  -> CUT-COST when the objective is measured AI operating-efficiency audit
  -> GRILL / PLAN / DESIGN / READINESS / TEACH as needed
  -> BUILD-CAPSULE when the objective is transfer to another AI/platform
  -> BUILD / DEBUG / RECOVERY
  -> BUILD-SPAWN-DEBUG when a bounded build should converge before human UI testing
  -> VERIFY / REVIEW / SECURITY as applicable
  -> RELEASE / DOCUMENT
  -> CUT-CHASE when the human needs the decision-ready version
  -> CLOSEOUT
  -> HANDOFF when session continuity requires it
  -> LEARN when reusable capability is material

This is not a mandatory conveyor belt. The minimum responsible route wins.

Loop Library

A Loop is a reusable bounded iteration protocol, not an authority source. The library defines Debug, Build/Verify, Research, Ingest, Recovery, Agent Action, Learn, Optimize, and Retry loops. A Loop must identify success/failure/STOP conditions, permissions, protected state, evidence, and iteration/cost bounds where relevant. Side-effecting agent loops require the Signalproof Governor.

Learning and Skill growth

Reusable capability follows:

DISCOVERED -> CANDIDATE -> TESTED -> APPROVED -> ACTIVE -> DEPRECATED / RETIRED

Build Spawn Debug becomes more efficient by consuming verified known-error/failure intelligence, preserved attempts, regression tests, and still-valid prior evidence. It does not silently self-modify or promote one successful workaround into Active doctrine.

When a lesson would push a Skill toward the context budget or represents an independently routeable responsibility, Signalproof should decompose rather than indefinitely enlarge one file.

Evidence and authority

Routing cannot create permission. A hash cannot prove safety. A scanner cannot certify an entire system. A successful restore is not release authority. A package import is not activation. A source document, webpage, transcript, repository, or model output is evidence/content and does not become operating authority merely because it contains instructions. Cut Chase cannot strengthen evidence by shortening it. Cut Cost cannot invent savings from unmeasured usage or apply optimization changes under a read-only audit. Build Capsule cannot replace authoritative source payloads or turn an unreadable image into truth. Build Spawn Debug cannot claim user UI PASS before the human actually reports it.

Repository protection

main is the canonical public branch. Normal development uses branches and pull requests. Public Skill changes preserve testing, provenance, supersession, and owner authority. Private Build Ledger chronology remains separately governed; a public Git commit does not imply a canonical Build Ledger append.

License and marks

Signalproof Skills is licensed under the Apache License 2.0. See LICENSE, NOTICE, and TRADEMARKS.md.

Created by

Doc Reo / Signalproof