The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from a local file (.artifacts/HANDOFF.md) to populate Obsidian notes, creating a surface for potential instruction injection.
Ingestion points: Content is read from .artifacts/HANDOFF.md as described in references/handoff.md.
Boundary markers: The agent is explicitly instructed to treat handoff data as claims to be verified against the current conversation or repository state rather than as authoritative commands.
Capability inventory: The skill uses Obsidian MCP tools for writing, patching, and moving notes, and executes a local Python script for configuration.
Sanitization: Note content is filtered through audience and reference discipline rules, and filenames are sanitized to remove problematic characters before creation.
[DYNAMIC_EXECUTION]: The skill executes a bundled Python script to resolve project settings and manage its registry.
Evidence: references/mapping.md and references/bootstrap.md direct the agent to execute scripts/registry.py using python3.
[COMMAND_EXECUTION]: The bundled registry script interacts with the system environment by executing git commands.
Evidence: scripts/registry.py calls subprocess.run to execute git worktree list --porcelain for repository root resolution.
[COMMAND_EXECUTION]: Skill instructions for project bootstrapping assemble a CLI command using string interpolation of user-provided values.
Evidence: references/bootstrap.md provides a template: python3 <this-skill>/scripts/registry.py add --name "<name>" --path "<obsidian-path>" --tags "<tags>", which could be vulnerable to shell injection if inputs are not properly escaped.

wrap-up