[trefeon]

Changes the repo's goal from a scanner to the agentic-AI work standard: the deterministic engine stays (569 rules, 9 domains) and the reference corpus is now compiled into a portable workflow skill library that the engine keeps honest.

The workflow library: 13 agentic-* skills

Each skill owns exactly one decision, names the sibling to use instead, cites the corpus path behind every non-obvious claim, and passes the repo's own scanner at 0 errors.

skill	owns	not for
agentic-frame	framing a raw request: goal, acceptance criteria, scope boundary, ask-vs-proceed	sequencing (agentic-plan)
agentic-plan	execution shape: plan doc plus prototype, slices, contract-first, risk-scaled rigor	goal choice; evidence (agentic-verify)
agentic-delegate	work split across agents: slice design, briefs, isolation, integration review	long-run supervision (agentic-loops)
agentic-context	window engineering: budgets, progressive disclosure, files as memory, compaction	durable memory (agentic-memory)
agentic-verify	proving work: evidence classes, falsification, failing-before/passing-after	diagnosis (agentic-debug)
agentic-debug	wrong or unexplained behavior: reproduce, narrow, bisect, root cause, stop rules	verification; performance tuning
agentic-review	judging a diff: review axes, receiving review, merge gate	git mechanics (agentic-git)
agentic-git	workspace and history: worktrees, branches, commit subjects, PR flow	review content; release steps
agentic-secure	secrets and supply chain: approval gates, pre-push scanning, untrusted content, pinning	app-security review; CI config rules
agentic-memory	durable knowledge: what to save, project vs global, recall, staleness, handoff	within-window context (agentic-context)
agentic-skill	authoring instruction artifacts: frontmatter, triggers, budgets, testing, non-overlap	auditing slop (slop-audit-skill)
agentic-harness	the agent's environment: tool design, permissions, hosted runtimes, observability	delegation; secrets policy
agentic-loops	operating over time: budgets, stop conditions, drift, escalation, eval loops	one-shot fan-out (agentic-delegate)

Compiled from the corpus's own workflow material: superpowers (15 skills), agent-skills (25), agent-skills-for-context-engineering (24), anthropics-skills, agentskills spec, DevOps-Security-Agent-Skills, fast-agent, hermes-agent-self-evolution, obsidian-skills. The corpus ships 60+ overlapping workflow skills; the library is the compiled, non-overlapping subset, with each source's prescriptions distilled rather than copied and the thin spots reported instead of padded.

The audit family (slop-audit*, 10 packs) is unchanged in intent: it judges a finished artifact against the rules DB, while agentic-* governs the work before the artifact exists. The zero-slop router now routes both families.

Citation layer fix

The corpus is cloned flat as reference/<manifest-name>/, but the packs cited 23 owner/repo forms that do not resolve on disk (reference/obra/superpowers:..., reference/jakubkrehel/skills:...) plus bare tail citations after a ; separator. scripts/fix-citations.mjs maps every citation through scripts/reference-corpus.json (owner/repo to manifest name) and normalizes the tail form. Result: 112 two-segment citations, 0 unresolved, with each tail verified against the corpus path. The script stays next to sync-reference.mjs because the corpus refreshes.

Install truth

• The README claimed npx -y zero-slop@latest scan .. The unscoped zero-slop name on npm belongs to another project (manavmishra/ZeroSlop), so that command executed a stranger's package. Every install command in the README, docs/PUBLIC_ADOPTION.md, docs/SHIP_PLAN.md and the zero-slop router now names @zero-slop/cli, and the README states the CLI is not published yet with the clone path to run today.
• Added a root pnpm build (core then CLI), which is what the clone instructions use.
• The npm version badge was removed until the package exists; the publish path is .github/workflows/publish.yml on a v* tag.
• Skills are installable today and this was verified live: npx skills use trefeon/zero-slop@slop-audit fetched the pack from the public repo and rendered its SKILL.md.

Docs

README (new tagline, install, workflow-library table), docs/MULTI_AGENT_SKILLS.md (two families, tree), docs/SHIP_PLAN.md (M1.8 checklist, state line, the fixed ship-acceptance command), docs/ZERO_SLOP_FRAMEWORK.md (goal, 84-repo evidence base, M1.8 milestone), root package.json description.

Verification

• Every skill gates at 0 errors individually and gate skills/ covers the whole tree; gate docs/ and gate README.md also 0 errors.
• pnpm validate:rules healthy, pnpm typecheck clean, core and CLI test suites pass.
• Full-repo gate: 0 error-tier findings.
• Public-repo scanners: gitleaks clean; trufflehog clean with sensitivity proven by a high-entropy control.

Not in this PR

npm publish (needs npm auth or trusted-publisher setup), the GitHub Action, the MCP server.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.