v4.9.0 — Trust narrative, new evals, plugin fix, GitHub Pages

What's new since v4.8.1

Trust and integrity reframe

• Root README now leads with trust and integrity verification, not just version tracking. Three audience hooks: authors, teams, trust.
• Comparison table gains an Integrity column showing SHA-256 hash support.
• "Trust and audit" section rewritten as "Trust and integrity" with concrete validate.sh walkthrough, team sharing workflow, and honest "what it does NOT do" disclosure.
• ProSkills.md badge (8.41/10) and ClawHub installs badge (333) added.

New use case walkthroughs (bundle README)

• Verifying a downloaded skill — step-by-step: check manifest, run hash check, read changelog, check source. Explains what a hash failure means.
• Sharing a skill across a team — canonical source, derived copies, deployment tracking, "are we on the latest?" workflow.

New evals (26 core, 30 total)

• #23: Settings UI .skill ZIP round-trip — verifies export/import preserves integrity
• #24: GitHub Copilot / VS Code bootstrap — .github/skills/ workflow
• #25: Cursor bootstrap — .cursor/skills/ workflow
• #26: Cowork filesystem persistence — session close without handoff

Plugin system fix

• Added marketplace.json so claude plugin marketplace add and claude plugin install work end-to-end.
• Added explicit skills path to plugin.json for reliable discovery.

GitHub Pages

• skillprovenance.dev — single-page landing site with teal/crimson palette, OG image, credibility signals (ProSkills 8.41/10, ClawHub 333 installs), and install instructions for Claude Code, Settings UI, ClawHub, Codex, and Gemini CLI.

Platform updates

• Gemini CLI: partial → pass (loading, bootstrap, and Gems workflow validated)
• Perplexity Computer: partial → pass

Bundle status

• Bundle version: 4.9.0
• SKILL.md: v16 (no definition changes)
• evals.json: v10 (26 core evals)
• README.md: v18 (use case walkthroughs added)
• All hashes verified via validate.sh

Install

Claude Code plugin:

claude plugin marketplace add snapsynapse/skill-provenance
claude plugin install skill-provenance@snapsynapse-skill-provenance

Claude Settings UI: download skill-provenance.skill below → Settings → Skills → Add Skill

ClawHub: clawhub install skill-provenance

Manual: clone the repo and use skill-provenance/ directly

v4.8.1 — PAICE.work PBC ownership + README restructure

Ownership

Skill Provenance is now a PAICE.work project. PAICE.work PBC is a public benefit corporation building infrastructure for productive collaboration between humans and autonomous agents.

• Copyright and author fields updated to PAICE.work PBC across LICENSE, plugin.json, and SKILL.md
• MANIFEST.yaml hash updated, .skill ZIP regenerated
• GitHub repository remains at snapsynapse/skill-provenance

README Restructure

• Leads with one-line value prop and comparison table instead of burying them below the problem statement
• "When not to use this" moved to its own section for honesty and scannability
• Graceful Boundaries added to Related projects (cross-linking PAICE.work projects)

New: CLAUDE.md

Added project instructions file for Claude Code users: key files, commands, conventions, and post-edit checklist.

No skill definition changes

The skill behavior, evals, and manifest schema are unchanged from v4.8.0. This release updates metadata and documentation only.

v4.8.0 — Claude Code plugin + repo hardening

This release adds Claude Code plugin infrastructure so the repo can be installed directly as a plugin, and tightens up the repository for marketplace submission.

What's new since v4.7.2

Claude Code plugin support (4.8.0)

• Plugin manifest (.claude-plugin/plugin.json): the repo now doubles as a Claude Code plugin. Install with:

  /plugin marketplace add snapsynapse/skill-provenance
  /plugin install skill-provenance@snapsynapse-skill-provenance
  

• Skills symlink (skills/skill-provenance → ../skill-provenance/): bridges the plugin system's discovery path without restructuring the existing bundle.
• README updated: plugin install is now the first listed method under Quick Install. Repo structure diagram reflects the new files.

Self-provenance (4.7.3)

• SKILL.md now exemplifies its own convention: added skill_bundle, file_role, version, version_date, previous_version, and change_summary to its own metadata block.
• Author attribution changed from "Snap Synapse" to "Sam Rogers (snapsynapse.com)".

Repo hardening

• GitHub Actions CI (validate.yml): automated hash verification and package build testing on every push and PR.
• SECURITY.md: vulnerability reporting policy with scoped in/out guidance.
• Archive gitignored: historical material removed from tracking. Files remain in git history for reference.

Bundle status

• Bundle version: 4.8.0
• SKILL.md: v15
• evals.json: v9 (22 core evals) — unchanged
• evals-distribution.json: v1 (4 supplemental evals) — unchanged
• All hashes verified via validate.sh

Install

Claude Code plugin: /plugin marketplace add snapsynapse/skill-provenance
Claude Settings UI: download skill-provenance.skill below → Settings → Skills → Add Skill
ClawHub: clawhub install skill-provenance
Manual: clone the repo and use skill-provenance/ directly

v4.7.2 - streamline packaging and split changelog history

Since 4.6.1, Skill Provenance has improved portability, publishing clarity, and package hygiene. Newer releases added author/source attribution metadata, a clearer three-state model for canonical bundles vs. strict-platform copies vs. registry packages, a zero-dependency package.sh helper for strict and ClawHub builds, and a supplemental eval suite for distribution/package scenarios.

This update also changes changelog handling to keep the bundle lightweight: skill-provenance/CHANGELOG.md now carries only the five most recent entries, while the full append-only release history lives in the repo root. Generated artifacts now default to build/ inside the project, with the ClawHub upload folder at build/clawhub/skill-provenance/.

v4.6.1 — Deployment Tracking, Trust Checks, and Expanded Evals

Overview

4.6.1 is the next published release after 4.5.0, so this release includes all changes since 4.5.0:

• stronger discovery-oriented SKILL.md wording
• Perplexity Computer testing and .zip upload guidance
• optional deployment metadata for cross-surface tracking
• trust and audit guidance for downloaded or reused bundles
• expanded eval coverage for deployment drift, trust checks, and package integrity

Highlights

Better discovery and portability

• Updated SKILL.md description to be more trigger-oriented for skill loaders
• Kept SKILL.md under the 500-line guidance
• Added .zip upload guidance for platforms that do not accept .skill
• Documented Perplexity Computer as a tested partial-compatibility surface

Cross-surface deployment tracking

• Added optional deployments guidance to the manifest model
• Clarified how to track local bundles, uploaded bundles, and API-deployed copies without replacing platform-native version systems
• Added guidance for .agents/skills/ as a neutral cross-client install path

Trust and audit improvements

• Added trust/audit framing to help verify bundles before reinstalling or reusing them
• Strengthened guidance around hash verification, changelog review, and stale deployed copies

Expanded eval suite

The eval suite now covers 22 scenarios total, including new coverage for:

• deployment drift after local changes
• multi-surface install targets
• trust checks for untrusted bundles
• redeploy events
• deployment metadata conflicts
• hash mismatches
• packaged subset vs canonical bundle handling

Files updated

• skill-provenance/SKILL.md
• skill-provenance/README.md
• skill-provenance/evals.json
• skill-provenance/MANIFEST.yaml
• skill-provenance/CHANGELOG.md
• README.md
• skill-provenance.skill

Validation

• Bundle version: 4.6.1
• Evals: 22 total
• Hashes refreshed and verified successfully with validate.sh

v4.5.0 — API surface, cross-surface fragmentation, dependencies field

This release documents the expanded Agent Skills ecosystem (30+ adopters, Claude API, Agent SDK) and positions cross-surface fragmentation as a core value proposition. It also adds a dependencies field to the manifest spec and cleans up terminology.

What changed:

• Expanded frontmatter constraint table to distinguish Claude Chat, Claude Code (with extensions), Claude API, and 30+ agentskills.io adopters
• Added Claude API and Agent SDK documentation to cross-surface section (epoch-timestamp versioning, skill isolation between surfaces)
• Promoted cross-surface fragmentation as headline pitch in root README
• Added dependencies field to manifest spec (empty-list default, omit if none)
• Renamed frontmatter_mode: claude to frontmatter_mode: metadata throughout (reflects spec-standard terminology)
• Trimmed Gemini Gems workflow section to single-paragraph note with eval reference
• Added skills-ref validation tool mention to references
• Documented metadata.version as spec-blessed, progressive disclosure model, and ${CLAUDE_SKILL_DIR} variable
• Added Skills API row to comparison table and compatible_with block to MANIFEST.yaml
• Added evals 14 (API surface awareness) and 15 (dependencies field usage) — now 15 evals total

Why this release:

The skills ecosystem has grown significantly since the last release. Skills now span the Claude API, Agent SDK, and 30+ third-party tools — but they don't sync between surfaces. This update documents that landscape and makes cross-surface fragmentation an explicit part of the project's pitch rather than an afterthought.

Files changed:

• SKILL.md, README.md (bundle and root), evals.json, MANIFEST.yaml, CHANGELOG.md

Validation: All hashes verified via validate.sh; .skill ZIP rebuilt from updated contents.

v4.3.0 — portability trim

This release makes skill-provenance lighter and more portable across models and surfaces by tightening the core protocol and downgrading surface-specific conveniences from defaults to optional artifacts.

What changed

• Simplified the core protocol around the truly portable bundle artifacts:
  MANIFEST.yaml, CHANGELOG.md, and the tracked bundle files.
• Made handoff.md optional instead of a default bundle requirement.
  Handoff notes are now recommended mainly for stateless chat transitions.
• Changed git commit output to inline-by-default.
  git_commit.txt is now an optional convenience artifact rather than a standard expected output.
• Reworked compatibility metadata to describe target surfaces and capabilities instead of pinning the bundle to a single exact model/surface tuple.
• Reduced Claude-first wording in the documentation and added clearer Codex loading guidance.
• Updated evals so portability-oriented behavior is the default:
  chat handoff is still validated where appropriate, but handoff.md and git_commit.txt are no longer universal success criteria.

Why this release

The goal of this update is to keep the bundle maximally portable and lightweight across Claude, Codex, Gemini CLI, and similar agents without carrying extra workflow assumptions from any one surface.

Files changed

• SKILL.md
  Refined the protocol so transient artifacts are optional and compatibility guidance is capability-based.
• README.md
  Reworked user guidance for a more agent-agnostic, lightweight workflow.
• evals.json
  Updated expectations to match the new portability-first protocol.
• MANIFEST.yaml
  Bumped bundle version and updated compatibility metadata.
• CHANGELOG.md
  Added the 4.3.0 release entry.

Validation

• ./validate.sh passes with all hashes verified.
• Rebuilt skill-provenance.skill from the updated bundle.

v4.2.1 — Semver, portability, ClawHub readiness

What's new

• Semver adoption: bundle_version now uses semver (4.2.1). Per-file versions remain integers. Retroactively mapped all changelog entries to semver, aligned with git tag v4.0.0.
• New repo files: AGENTS.md (agent guide), CONTRIBUTING.md (contribution guide), compatibility-report issue template.
• Root README rewrite: Value statement, before/after comparison, platform support matrix, comparison table (git tags vs filename suffixes vs Skillman vs Skill Provenance), and "when not to use this" guidance.
• validate.sh portability: Added sha256sum fallback for Linux alongside shasum (macOS).
• Related projects: Added pi0/skillman (JS/TS) alongside chrisvoncsefalvay/skillman (Python).
• Research reference: Added Xu & Yan (2026) agent skills survey paper.
• ClawHub readiness: Added ClawHub install path to README, metaskill terminology throughout.

Platform support

Platform	Status
Claude (Chat, Code, Cowork)	Pass
Codex (OpenAI)	Pass
Gemini CLI (Google)	Partial
GitHub Copilot	Untested
Cursor	Untested

Install

Claude Settings UI: Download skill-provenance.skill below and import via Settings → Skills → Add Skill.

Claude Code / Codex / Gemini CLI: Use the skill-provenance/ directory directly.

ClawHub: clawhub install skill-provenance

v4.0.0 — Gemini CLI support, validation tooling & cross-platform portability

What's new in v4

This release extends Skill Provenance to run cleanly on Gemini CLI, hardens the validation tooling, and ships the bundle in frontmatter_mode: minimal for maximum portability across all platforms.

Highlights

• Gemini CLI support — SKILL.md updated with Gemini CLI in the frontmatter constraint table and cross-platform considerations (three-tier discovery, install paths, management commands)
• • validate.sh — new local hash verification script; hardened for portable update-mode rewrites and locale-stable hash verification against MANIFEST.yaml
• • Expanded evals — 5 new eval scenarios (evals 9–13): Gemini CLI compatibility, Gemini Gems prompt extraction, git commit message generation, handoff with per-file changes, local hash validation. Now 13 evals total
• • Session close protocol — added git commit message generation (step 6) and required per-file handoff summaries
• • Minimal frontmatter mode — the bundle now ships in frontmatter_mode: minimal so the same source works across Claude, Codex, and Gemini CLI without modification
• • README updates — Gemini CLI quick-start, Gemini Gems workflow guidance, local hash validation section, and clarified that the checked-in skill-provenance/ directory is the canonical cross-platform source bundle

Install

Download skill-provenance.skill and install it:

Claude (web): Profile → Settings → Skills → Add Skill → select the file

Claude Code / Codex / Gemini CLI: use the skill-provenance/ directory directly (checked into this repo)

Platform compatibility

Platform	Status
Claude (chat)	✅
Claude Code	✅
Codex (OpenAI)	✅
Gemini CLI (Google)	✅ new in v4
GitHub Copilot	✅

Full changelog

See skill-provenance/CHANGELOG.md for the complete history going back to v1.