The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied knowledge base files (Markdown, wiki links) to perform reorganization and linting.
Ingestion points: Reads files from workspace.knowledge_root and workspace.project_root as specified in SKILL.md and skill.yaml.
Boundary markers: The SKILL.md execution boundary states that the skill will not bypass protections, automatically send messages, or force-end user processes.
Capability inventory: Performs filesystem operations including reading, moving, and updating references within the project root.
Sanitization: Relies on natural language analysis for deduplication and link updating; no explicit code-level sanitization against malicious instructions embedded in KB content is present.
[COMMAND_EXECUTION]: The skill includes a helper script scripts/profile_store.py to manage local configuration and persistence.
Evidence: The script performs file I/O on specific local profile paths such as ~/.config/agent-skills/profile.json to persist user, brand, and workspace metadata.
Safety features: The script implements a SENSITIVE_PARTS blacklist that explicitly blocks attempts to persist credentials, tokens, secrets, or passwords into the profile.
[EXTERNAL_DOWNLOADS]: The skill declares dependencies on other vendor resources for its full workflow.
Evidence: The SKILL.md frontmatter lists lov-branding-consistency as a dependency for reviewing public-facing text.

lov-kb-organize