feat: add AI agent skills inventory scanning support

[raysubham]

Summary

Adds a detector that inventories Agent Skills installed for AI coding agents, at both global and per-project scope. It covers Claude Code skills and skills.sh-managed installs (including plugin-provided skills).

For each discovered skill the scan records:

• Metadata: skill name, agent, scope, and source
• A SHA-256 hash of the skill's SKILL.md (for drift detection)
• skills.sh provenance from lock files (source, ref, install/update timestamps)
• A stat-only census of the skill folder (file counts and sizes — contents are not read)

Lock entries with no corresponding folder on disk are dropped — a lock entry is not an install, so the inventory is limited to on-disk skills.

Privacy

• Skill file contents are never read beyond SKILL.md, and are never transmitted.
• Only a single SHA-256 hash of each SKILL.md is recorded — never its contents.
• For local skill sources, the on-disk source path is never serialized; only the lock alias is recorded.

Notes

• SKILL.md discovery is case-sensitive (exact SKILL.md).
• The ~/.claude.json project-registry discovery is shared with the MCP detector.
• Enabled by default.

Testing

• go build ./..., go vet ./..., and gofmt are clean
• go test ./... passes for the affected packages

[Copilot]

Pull request overview

Adds a new filesystem-only detector that inventories “agent skills” (directories containing SKILL.md) across multiple AI coding agents and skills.sh-managed installs, and wires the results through community scan output, enterprise telemetry payloads, and both Pretty/HTML renderers.

Changes:

• Introduces SkillsDetector with bounded discovery (global + project + optional home-walk), SKILL.md frontmatter parsing + hash, stat-only folder census, and optional skills.sh lock provenance enrichment.
• Plumbs new model.AgentSkill / model.AgentSkillScanInfo into scan + telemetry payloads and exposes counts via summary/metrics and Pretty/HTML output.
• Shares Claude Code project-root discovery via new discoverClaudeProjects and adds a new TCC helper (WithinProtected) plus extensive unit tests.

Reviewed changes

Copilot reviewed 24 out of 24 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
SCAN_COVERAGE.md	Documents the new “AI Agent Skills” inventory coverage and privacy posture.
internal/telemetry/telemetry.go	Adds agent-skills collection phase to enterprise telemetry and includes results + metrics in the payload.
internal/telemetry/phase_deadline.go	Allocates a phase budget for agent_skills_scan.
internal/telemetry/collect_project_roots_test.go	Tests telemetry-side project-root mapping for the skills detector.
internal/tcc/tcc.go	Adds WithinProtected and hasDirPrefix for pre-access protected-tree checks.
internal/tcc/tcc_darwin_test.go	Adds darwin-only tests covering WithinProtected.
internal/scan/scanner.go	Runs the skills detector in community scans and wires results into ScanResult + summary.
internal/output/pretty.go	Renders an “AGENT SKILLS” section and summary count in Pretty output.
internal/output/pretty_test.go	Adds Pretty output tests for the new section’s “not scanned / none / populated” states.
internal/output/html.go	Adds Agent Skills summary card + section/table to HTML output.
internal/output/html_test.go	Adds HTML output tests for the new Agent Skills section states.
internal/model/model.go	Adds AgentSkill, AgentSkillScanInfo, and scan summary count wiring in ScanResult/Summary.
internal/featuregate/featuregate.go	Adds and enables FeatureAgentSkillsScan.
internal/featuregate/featuregate_test.go	Extends override test coverage to include the new feature gate.
internal/executor/mock.go	Extends mock executor to simulate symlink dir entries and symlink resolution errors.
internal/detector/skills.go	Implements the main skills discovery pipeline (roots, project discovery, symlink collapsing, caps).
internal/detector/skills_test.go	Comprehensive unit tests for skills parsing, census, lock parsing, symlink collapsing, and detector behavior.
internal/detector/skills_tcc_darwin_test.go	Darwin-only tests ensuring the detector avoids TCC-protected trees without triggering access.
internal/detector/skills_lock.go	Parses skills.sh lock files, sanitizes keys, and enriches discovered skills with provenance.
internal/detector/skills_homewalk_test.go	Tests home-walk project discovery and its caps/determinism.
internal/detector/skills_frontmatter.go	Parses YAML frontmatter from SKILL.md, hashes content, and scans body for load-time shell exec markers.
internal/detector/skills_census.go	Implements the stat-only directory census (counts/sizes/mtime/plugin manifest detection).
internal/detector/mcp.go	Switches MCP project discovery to reuse discoverClaudeProjects.
internal/detector/claudeprojects.go	New shared helper to read project roots from ~/.claude.json.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[raysubham]

This recover() is intentional, and it's the one spot the skills detector deliberately diverges from the "detectors return values; telemetry.Run handles panics" convention. Two reasons specific to this detector:

1. Detect returns (skills, info), where info is the "scan ran" sentinel. A nil AgentSkillScan means "no information," which the backend treats differently from "scan ran, found N." The local recover guarantees a non-nil info (with the partial inventory) even on a mid-walk panic.

2. Blast radius. Without it, a panic here propagates to telemetry.Run's top-level recover (telemetry.go:335), which marks the entire run failed and uploads no payload — discarding every other detector's results (IDEs, extensions, CLI tools, npmrc/pip audits, …) that already ran this pass. Containing it here keeps a bug in this new, default-on, most-complex detector from failing the whole telemetry run.

The failure isn't swallowed: the handler records the panic via addError(info, …) and sets info.Truncated = true, so the scan is surfaced as partial/non-authoritative (the backend then suppresses deletions for it) rather than silently succeeding. Given that, we'd prefer to keep it. Happy to revisit if we later add per-phase panic isolation in telemetry.Run — at that point this could move up to the phase runner and share the pattern with the other detectors.