The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from repository files to identify commands and package layouts, creating a potential surface for indirect prompt injection if those files are maliciously crafted.
Ingestion points: Processes package.json scripts, lockfiles (pnpm-lock.yaml, yarn.lock, etc.), and task configuration files (Makefile, nx.json, turbo.json).
Boundary markers: The skill does not employ specific delimiters or "ignore" instructions when interpolating discovered facts into generated plans.
Capability inventory: Capabilities are limited to directory creation and file writing (writing plans) via the setup-modules.mjs script.
Sanitization: Discovered strings are included in plans without explicit sanitization, though the skill does not execute these strings directly.
[DYNAMIC_EXECUTION]: The script scripts/setup-modules.mjs uses dynamic import() to load logic from the lodestar-setup skill. This is a form of code loading, but it is restricted to the local filesystem at a fixed relative path (../../lodestar-setup/scripts/) using hardcoded module names, which mitigates the risk of arbitrary path injection.
[EXTERNAL_DOWNLOADS]: The skill references a requirement for the lodestar-setup skill and provides an installation command for a suite of skills from the vendor's repository (ilancohen/lodestar-skills). This is documented as standard functionality for managing the tool's dependencies.

lodestar-plan