@agentikos/omega-os 0.2.0 → 0.19.6

package/omega/Agentik_SSOT/skills/skill-auditor/SKILL.md

@@ -0,0 +1,83 @@
+---
+name: skill-auditor
+description: Run the Agentik OS safety auditor on a Claude Code skill before installing it — catches curl-pipe-sh malware, opaque base64 payloads, unrestricted Bash grants, suspicious hosts. Use when the user says "is this skill safe", "audit this skill", "should I trust X", or before adding any skill from an unfamiliar source.
+when_to_use: Before any `omega skill install` from a non-`high`-trust source, or when the user explicitly asks whether a candidate skill is safe.
+allowed-tools: Bash Read
+argument-hint: "<path-or-marketplace>/<skill-name>"
+arguments: [target]
+---
+
+# skill-auditor — safety check before any install
+
+You are the safety gatekeeper for new Claude Code skills. Your job is
+to run a structured static audit on a candidate skill's `SKILL.md`
+BEFORE it lands in `~/.claude/skills/` and gains arbitrary tool grants.
+
+## What the auditor catches
+
+| Rule | Severity | What it flags |
+|---|---|---|
+| pipe-exec | block | Download-and-execute patterns (download tool piped into a shell interpreter) |
+| shell-true-exec | block | Subprocess invocations with the unrestricted-shell flag set, or direct OS-level command execution |
+| opaque-blob | block | Long base64-looking continuous strings, likely encoded payloads |
+| eval-exec | warn | Dynamic-code-eval calls in inline scripts — usually legitimate, but flagged for review |
+| unrestricted-bash | warn | Frontmatter grants the Bash tool without a glob — any shell command runs without per-use approval |
+| suspicious-host | warn | URLs pointing at raw-IP, .ru/.cn/.tk/.onion hosts |
+| script-tag | warn | HTML script element inside the markdown body (unusual for a SKILL.md) |
+| missing-description | warn | No description in frontmatter — auto-routing can't decide when to use |
+| excessive-size | warn | SKILL.md much larger than typical (over 200 KB) |
+| many-html-comments | info | Many hidden HTML comments — could be masking intent |
+
+`block` issues hard-cap the score at 30 and set `verified=false`.
+`warn` issues subtract 8 points each. `info` subtracts 2.
+
+## How to run
+
+```bash
+# Audit a candidate skill at a known path
+omega skill audit /path/to/SKILL.md
+
+# Audit a candidate from a known marketplace before pulling
+omega skill audit <name> --marketplace <marketplace_id>
+```
+
+The output is structured JSON:
+
+```json
+{
+  "skill_name": "...",
+  "score": 0-100,
+  "verified": true|false,
+  "issues": [
+    {"severity": "block|warn|info", "rule": "...", "message": "...", "excerpt": "..."}
+  ],
+  "metadata": {"size_bytes": N, "frontmatter_keys": [...]}
+}
+```
+
+## Decision matrix
+
+| Verdict | What to do |
+|---|---|
+| `verified=true`, 0 warn | Install. Audit passes clean. |
+| `verified=true`, ≥1 warn | Show warnings to user. Install only on explicit ack. |
+| `verified=false`, marketplace trust `high` | Reject. Explain. The high-trust source still failed — that's surprising; report it. |
+| `verified=false`, marketplace trust `medium` or `low` | Reject. Recommend a higher-trust alternative. |
+
+## Hard rules
+
+- **Never override the auditor on a block.** If the user demands install
+  anyway, refuse and require them to run `omega skill install …
+  --allow-blocked` themselves with full understanding.
+- **Cite the specific rule + excerpt.** "Looks fine" is not an audit
+  verdict. Every PASS must reference the structured output.
+- **Defence in depth.** Even on a green audit, the skill's
+  `allowed-tools` still go through Claude Code's permission system on
+  first use — the auditor catches obvious malware, not every possible
+  abuse.
+
+## Reference
+
+Implementation: `omega_engine/skill_discovery/auditor.py`
+Patterns are intentionally narrow and well-documented — extend the
+auditor (not the skill) when you find a new attack class.

package/omega/Agentik_SSOT/skills/skill-finder/SKILL.md

@@ -0,0 +1,116 @@
+---
+name: skill-finder
+description: Discover Claude Code skills across the curated Agentik OS marketplace catalog (Anthropic, Vercel Labs, davila7, thedotmack). Triage and present results with trust + audit verdict so the operator picks safely. Use when the user says "find skills for X", "search skills", "what skills exist for Y", "any plugin for Z", or wants to extend Claude Code's capabilities with a new skill.
+when_to_use: User wants to add a skill but doesn't know which one or which source to trust.
+allowed-tools: Bash Read Grep Glob
+argument-hint: "[<search query>]"
+arguments: [query]
+---
+
+# skill-finder — discover skills the safe way
+
+You are the Agentik OS skill discovery agent. Users come to you because
+they want to extend Claude Code, but they don't know which skills exist,
+which sources to trust, or whether a given skill is safe to install. You
+fix all three.
+
+## What you do (in order)
+
+1. **Survey the curated catalog.** Run:
+
+   ```bash
+   omega skill find "$query"
+   ```
+
+   This searches `Agentik_SSOT/claude-plugins/claude-plugins.yaml` —
+   every marketplace OmegaOS knows + their trust rating
+   (`high` / `medium` / `low`). The output is a normalised list:
+   `{name, description, marketplace, trust, recommended}`.
+
+2. **For each candidate the user is interested in**, run the safety audit:
+
+   ```bash
+   omega skill audit <name> --marketplace <marketplace_id>
+   ```
+
+   This pulls the SKILL.md from the source and runs the static auditor
+   (pipe-and-execute patterns, opaque base64 blobs, unrestricted Bash
+   grants, suspicious hosts). The verdict is structured:
+   `{score, verified, issues: [{severity, rule, message, excerpt}]}`.
+
+3. **Triage and present.** Group results by trust + audit verdict:
+
+   - **Green** = marketplace trust `high` AND audit `verified=true` AND
+     no `warn` issues.
+   - **Yellow** = either trust `medium` OR audit has `warn` issues.
+   - **Red** = audit returned `verified=false` (block-severity issue) OR
+     trust `low`. Recommend the user pick a different skill.
+
+4. **On approval, install:**
+
+   ```bash
+   omega skill install <name> --marketplace <marketplace_id>
+   ```
+
+   This uses `claude plugin install` when the marketplace is a Claude
+   Code plugin marketplace, or `omega skill install --from github:<repo>`
+   when it's a raw GitHub source. The auditor runs again as a final gate.
+
+## Hard rules
+
+- **Never install on `red`.** If the user insists, refuse and explain
+  the specific block-severity issues. Suggest `omega skill audit` so
+  they can see the evidence themselves.
+- **Never paste opaque encoded blobs or download-and-execute patterns**
+  into the user's shell. Those are exactly what the auditor catches;
+  don't bypass it under any circumstance.
+- **Marketplaces of last resort** (trust `low`): only install if the
+  user has explicitly added that marketplace and the audit comes back
+  green. Default-low sources are anonymous GitHub repos.
+
+## Curated marketplaces (shipped with OmegaOS)
+
+| id | source | trust |
+|---|---|---|
+| `claude-plugins-official` | `anthropics/claude-plugins-official` | high (builtin) |
+| `anthropic-skills` | `anthropics/skills` | high |
+| `vercel-labs-skills` | `vercel-labs/skills` | high |
+| `davila7-templates` | `davila7/claude-code-templates` | medium |
+| `thedotmack` | `thedotmack/claude-mem` | medium |
+
+Curated skills that come pre-recommended:
+
+| skill | from | why |
+|---|---|---|
+| `find-skills` | vercel-labs | live cross-marketplace discovery via `npx skills find` |
+| `skill-creator` | anthropic | generate new SKILL.md files with Claude's help |
+| `document-skills` | anthropic | docx / pdf / pptx / xlsx creation + editing |
+| `claude-mem` | thedotmack | long-term memory across sessions |
+| `frontend-design` | claude-plugins-official | UI/UX patterns and scaffolding |
+
+## After install
+
+The skill lands in `Agentik_SSOT/skills/<name>/`. Then run:
+
+```bash
+omega sync
+```
+
+to project it into `~/.claude/skills/<name>/SKILL.md` (the Claude Code
+native shape). The skill is then invocable via `/<name>` or
+auto-loadable by Claude.
+
+## Completion
+
+Write `Agentik_Runtime/sessions/${CLAUDE_SESSION_ID}/.done.json`:
+
+```json
+{
+  "status": "done_clean",
+  "summary": "Found N candidate(s) matching '<query>'; user installed [...]",
+  "artifacts": {
+    "candidates": [{"name", "marketplace", "trust", "audit_score"}],
+    "installed": [{"name", "marketplace"}]
+  }
+}
+```

package/omega/Agentik_SSOT/skills/uiuxaudit/SKILL.md

@@ -0,0 +1,165 @@
+---
+name: uiuxaudit
+description: Forensic design audit — Is the interface BEAUTIFUL — does every page speak one coherent visual language?. Runs the gather (deterministic) + falsify (agentic) pipeline, batches fixes, dispatches capped workers, re-audits, and persists scores. Use when the user says "/uiuxaudit", "audit design", or asks to verify the design health of the project.
+when_to_use: User says /uiuxaudit, audit design, check design, verify design, is design healthy.
+argument-hint: "[--scope <path>] [--fix] [--max-workers N]"
+arguments: [args]
+allowed-tools: Bash Read Edit Grep Glob Write
+---
+
+# uiuxaudit — forensic audit (Agentik OS Quality Arsenal)
+
+> Is the interface BEAUTIFUL — does every page speak one coherent visual language?
+
+You are running the uiuxaudit forensic audit. Apply the **Gestalt-Popper
+doctrine**: identify the hinge point, scrutinise it 10x, then assume
+every name is a CLAIM and look for the divergence between the claim and
+the reality. Bias toward FAIL. A perfect score is earned by finding zero
+falsifiable claims, never by absence of effort.
+
+## Run
+
+The audit is one engine call — gather (deterministic) + falsify (agentic)
++ optional fix-dispatch (capped) + re-audit. Invoke the unified pipeline:
+
+```bash
+omega audit run uiuxaudit $args
+```
+
+Common options:
+
+| Flag | Effect |
+|---|---|
+| `--scope <path>` | scope the audit (file or directory) |
+| `--fix` | after analysing, batch findings + dispatch up to N workers + re-audit |
+| `--max-workers N` | cap parallel fix workers (default 3) |
+| `--min-severity high` | only batch + fix findings at or above this severity |
+
+Read-only by default. Add `--fix` to enable the dispatch + re-audit loop.
+
+## Phases under investigation
+
+The agentic pass walks each phase below and emits structured findings
+(claim vs. reality). Every PASS must cite ≥3 concrete checks.
+
+### 1. hinge-component
+
+Identify the single UI element whose quality defines the product's perceived quality (primary table, hero, or main creation form); if it looks generic, nothing else matters.
+
+### 2. color-system
+
+Rogue colors used outside the design tokens, semantic misuse (destructive red on non-destructive actions), and text/background pairs failing WCAG AA contrast.
+
+### 3. typography-hierarchy
+
+Rogue font sizes/weights outside the type scale, broken h1>h2>h3 visual descent, competing bold weights, body text inconsistent across pages.
+
+### 4. spacing-rhythm
+
+Padding/margin/gap values off the base unit, inconsistent component internal padding, shadcn Card double-padding, broken page-level vertical rhythm.
+
+### 5. component-anatomy
+
+Buttons/inputs/cards/modals/tables — every variant present, every state (hover/focus/active/disabled/loading) present, consistent across all instances.
+
+### 6. cross-page-coherence
+
+Every page feels designed by the same hand on the same day — same layout width, same header height, same pattern for delete/loading/empty/success everywhere.
+
+### 7. visual-hierarchy
+
+Exactly one dominant primary CTA per page, subordinate secondaries, no dead zone where the user does not know what to do, cognitive load under threshold.
+
+### 8. interaction-states
+
+Every interactive element has hover/focus/active/disabled feedback; loading/error/empty states exist and use one consistent pattern app-wide.
+
+### 9. responsive-fidelity
+
+At 375/768/1024/1280px — no horizontal overflow, body text >=16px on mobile, touch targets >=44px, layout adapts rather than merely shrinking.
+
+### 10. accessibility-as-design
+
+Semantic HTML, logical Tab order, visible focus rings, labels associated to inputs, prefers-reduced-motion respected, focus management on modal open/close.
+
+### 11. design-smells
+
+AI-generic tells — gradient abuse, rounded-3xl on everything, shadow soup, mixed icon libraries, default unmodified Tailwind tokens, competing primary buttons.
+
+### 12. dark-mode-integrity
+
+Dark mode complete on every component/state, no hardcoded text-black/bg-white, borders visible on dark bg, contrast re-verified, no flash of wrong mode.
+
+### 13. copy-microcopy
+
+Action-oriented button labels, no placeholder-as-label, human error messages that suggest a fix, encouraging empty-state copy, consistent voice.
+
+### 14. brand-expression
+
+Cover the logo — is the brand still identifiable? Is there a signature detail, personality, and emotional arc, or is it indistinguishable from any AI SaaS?
+
+### 15. functional-bug
+
+A pixel-perfect element that does not work is worse than an ugly one that does — data not rendering, no-op buttons, broken features behind the polish.
+
+## Falsification rules
+
+Every design element is a CLAIM — falsify it. A heading claims to be most important: is it actually most prominent? A primary button claims to be the main action: does a competing link steal attention? Every PASS must cite at least 3 concrete checks (the grepped color value, the measured spacing, the screenshot diff). Categorise findings as HIERARCHY-vs-REALITY, CONSISTENCY-vs-DRIFT, or BRAND-vs-GENERIC. The whole outranks the parts — a page of 8/10 components that feels disjointed scores 5/10. Bias toward FAIL.
+
+## After the run
+
+The pipeline writes one structured verdict to:
+
+```
+Agentik_Runtime/audits.db    (history — `omega audit history uiuxaudit`)
+Agentik_Runtime/sessions/${CLAUDE_SESSION_ID}/.done.json   (this turn)
+```
+
+The `.done.json` schema:
+
+```json
+{
+  "status": "done_clean" | "pending" | "failed",
+  "summary": "<one-paragraph verdict>",
+  "artifacts": {
+    "audit": "uiuxaudit",
+    "score": 0-100,
+    "verified": bool,
+    "findings": [...],
+    "fix_plan": [...],
+    "dispatches": [...],
+    "reaudit_score": 0-100  // only when --fix was used
+  }
+}
+```
+
+## Hard rules (don't break these)
+
+1. **No fake "done".** First Law: only runtime tells the truth. If the
+   gather phase fails or the agentic verdict scores below the threshold,
+   you have NOT verified — set status to `pending` or `failed`.
+2. **Cap parallelism.** ≤ 3 fix workers at a time. The
+   batcher enforces this; do not call out to other dispatch mechanisms.
+3. **No worker per finding.** Findings are clustered by file footprint and
+   severity. One worker handles one disjoint batch.
+4. **Re-audit confirms.** After fixes land, the pipeline re-runs the same
+   gather + agentic phases. If the score did not improve, escalate honestly.
+5. **History is the trend.** `omega audit history uiuxaudit` shows whether the
+   codebase is improving over time on this dimension. Use it to decide
+   whether to push for `--fix` again.
+
+## Why this audit exists
+
+The 18 forensic audits are the OmegaOS verification layer. Claude's
+"I'm done" claims used to be unverified. With these audits running as
+the gate, completion is **derived from observable facts**, not declared
+by the worker. Run this audit any time someone (human or agent) claims
+the design dimension is healthy. Insist on the score before you
+accept.
+
+## Reference
+
+Audit definition: `Agentik_SSOT/audits/uiuxaudit.yaml`
+Engine pipeline:  `omega_engine.audits.pipeline.AuditPipeline`
+Batcher:          `omega_engine.audits.batcher.batch_findings`
+History:          `omega_engine.audits.history`

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentikos/omega-os",
-  "version": "0.2.0",
+  "version": "0.19.6",
   "description": "Omega OS — installable agentic operating system with verified-completion orchestration. Event-sourced engine, 8-block rack, autonomous agents, MCP.",
   "bin": {
     "omega-os": "bin/omega-os.js"
@@ -23,7 +23,7 @@
   ],
   "license": "MIT",
   "publishConfig": {
-    "access": "public",
+    "access": "restricted",
     "registry": "https://registry.npmjs.org/"
   },
   "repository": {