The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various git commands (status, add, commit, restore) and shell-based build/test commands (typecheck, test) defined in a configuration file (context.md).
Evidence: SKILL.md and references/execute.md detail the use of git for staging and committing changes, and the execution of build commands to verify fixes.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and act upon instructions (the "Suggested fix") contained within markdown files generated by a separate audit process. This represents a vulnerability surface where malicious instructions in the audit report could influence the agent's actions on the source code.
Ingestion points: NNN-<category>-<slug>.md action item files read during the Execute step (Step 3).
Boundary markers: The skill specifies sections (problem, fix, scope) but lacks explicit boundary markers or "ignore embedded instructions" warnings for the fix content itself.
Capability inventory: File system modification (writing fixes), git operations, and execution of build/test scripts.
Sanitization: The skill performs sanitization on commit messages (checking for newlines and control characters) but does not verify the safety of the code fixes suggested in the audit items.
[DYNAMIC_EXECUTION]: The skill uses dynamic module imports to load shared utilities from a sibling skill directory and executes build commands retrieved from an external context file.
Evidence: scripts/setup-modules.mjs uses import() to load logic from ../../lodestar-setup/scripts/. references/execute.md executes <typecheck> and <test> commands via the shell.

lodestar-fix