The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project manifests and external documentation which creates a surface for indirect prompt injection.
Ingestion points: The skill reads project-level files such as package.json and pnpm-lock.yaml, as well as external sources like official release notes and migration guides (SKILL.md, sections 2 and 3).
Boundary markers: There are no explicit delimiters or instruction-ignore wrappers specified for the ingested content.
Capability inventory: The skill has file-write access for updating manifests and lockfiles, and shell execution capabilities via its wrapper scripts.
Sanitization: No specific filtering or sanitization logic is described for the text strings extracted from manifests or external release notes.
[COMMAND_EXECUTION]: The skill relies on local shell command execution for its core functionality.
Evidence: The skill executes internal bash scripts check-prerequisites.sh and run-pnpm.sh.
Context: The run-pnpm.sh script executes pnpm commands inside an interactive shell (zsh -ic) to ensure the user's Socket Firewall (sfw) alias is utilized for all operations, which is a standard pattern for developer-focused tools.

safe-pnpm-updater