@mondoohq/skillcheck 0.3.4 → 0.4.0

package/README.md

@@ -27,6 +27,14 @@ npx @mondoohq/skillcheck            # scan every detected agent
 npx @mondoohq/skillcheck --verbose  # include content hashes and report URLs
 ```
 
+Agents also load skills and instructions from the project you open. Add `--project` to scan
+a checkout too: its skills (including `.skill` packages) and the instruction files agents
+read there, such as `AGENTS.md`, `CLAUDE.md`, `.cursor/rules` and Copilot instructions.
+
+```bash
+npx @mondoohq/skillcheck --project .
+```
+
 ### Stop risky skills from reaching your builds
 
 skillcheck exits **1** when it finds a critical or high-risk skill, so it works as a gate in
@@ -49,8 +57,9 @@ It fails open: a skill the database hasn't analyzed yet shows as clean and never
 If you publish skills, `validate` checks your repository against the
 [Agent Skills](https://agentskills.io/specification) and [AGENTS.md](https://agents.md/)
 specifications — catching broken frontmatter, bad skill names, oversized descriptions,
-and a missing `AGENTS.md` or marketplace manifest before your users do. It finds every
-`SKILL.md` in the repository, including skills nested in plugin directories.
+links to files the skill doesn't ship, and a missing `AGENTS.md` or marketplace manifest
+before your users do. It finds every `SKILL.md` in the repository, including skills nested
+in plugin directories and skills packaged as `.skill` archives.
 
 ```bash
 npx @mondoohq/skillcheck validate .                    # current directory
@@ -61,22 +70,26 @@ npx @mondoohq/skillcheck validate ./my-skills --json   # any path, JSON output
 Validating /path/to/repo
 
 Agent Skills spec (SKILL.md)
-  ✓ The repository contains at least one skill (a SKILL.md file)
-  ✓ Every SKILL.md starts with valid YAML frontmatter with correctly typed fields
+  ✓ Every skill is readable, with valid, correctly typed YAML frontmatter
   ✓ Every skill name is a lowercase slug (a-z0-9, single hyphens)
   ✓ Every skill name is 1-64 characters
-  ✓ Every skill name equals its directory name
   ✓ Every skill description is 1-1024 characters
+  ✓ Every file a skill refers to is bundled with it
+  ✓ Every skill name equals its directory name
   ✓ Every skill compatibility note is at most 500 characters
-Repository contract (agents.md + marketplace)
+  ! Every SKILL.md body is at most 500 lines (move detail into references) (warning)
+Repository contract (skills, agents.md, marketplace)
+  ✓ The repository contains at least one skill (SKILL.md or .skill package)
   ✓ A root AGENTS.md exists (agents.md convention)
   ✓ .claude-plugin/marketplace.json exists and lists plugins
 
-PASS 9 passed, 0 failed
+PASS 10 passed, 0 failed, 1 warning(s)
 ```
 
-It exits **1** if any check fails, so it drops straight into a skills repo's CI. The rules
-are an embedded [MQL](https://mondoo.com/docs/mql/home/) policy
+Each check has a severity. It exits **1** if a check fails; a low-severity check, like the
+500-line recommendation, is a warning that is shown but doesn't fail. That drops straight
+into a skills repo's CI. The rules, and their severities, are an embedded
+[MQL](https://mondoo.com/docs/mql/home/) policy
 ([`repo-contract.mql.yaml`](https://github.com/mondoohq/skillcheck/blob/main/internal/validate/policy/repo-contract.mql.yaml)),
 and because it is a standard cnspec bundle, `cnspec scan filesystem <repo> -f
 repo-contract.mql.yaml` runs the identical checks.
@@ -89,6 +102,10 @@ For each detected agent, skillcheck:
 2. Computes a SHA-256 content hash for each skill
 3. Looks the hash up in the [Mondoo skill database](https://mondoo.com/ai-agent-security/skills)
 4. Reports findings with severity, summary, and a link to the full security report
+5. Checks each skill against the [Agent Skills specification](https://agentskills.io/specification),
+   with the same rules as `validate`, and lists what it fails under the skill — such as a
+   name that isn't a valid slug, or images and references an installer didn't copy. These
+   are shown but never change the exit code: a malformed skill is not a malicious one.
 
 Only hashes are sent, never skill contents. See
 [supported agents](https://github.com/mondoohq/skillcheck/blob/main/docs/supported-agents.md)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mondoohq/skillcheck",
-  "version": "0.3.4",
+  "version": "0.4.0",
   "description": "AI agent skill security scanner. Detects locally installed agent skills, computes SHA-256 checksums, and queries the Mondoo AI Agent Security database for known threats.",
   "keywords": [
     "ai-security",
@@ -33,11 +33,11 @@
     "node": ">=18"
   },
   "optionalDependencies": {
-    "@mondoohq/skillcheck_darwin_amd64": "0.3.4",
-    "@mondoohq/skillcheck_darwin_arm64": "0.3.4",
-    "@mondoohq/skillcheck_linux_amd64": "0.3.4",
-    "@mondoohq/skillcheck_linux_arm64": "0.3.4",
-    "@mondoohq/skillcheck_windows_amd64": "0.3.4",
-    "@mondoohq/skillcheck_windows_arm64": "0.3.4"
+    "@mondoohq/skillcheck_darwin_amd64": "0.4.0",
+    "@mondoohq/skillcheck_darwin_arm64": "0.4.0",
+    "@mondoohq/skillcheck_linux_amd64": "0.4.0",
+    "@mondoohq/skillcheck_linux_arm64": "0.4.0",
+    "@mondoohq/skillcheck_windows_amd64": "0.4.0",
+    "@mondoohq/skillcheck_windows_arm64": "0.4.0"
   }
 }