The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute update and version commands for external CLI tools (codex, claude, pi, agy). While these are legitimate maintenance actions, they represent a significant capability surface.
Evidence: Sequential or parallel execution of codex update, claude update, pi update, and agy update commands in SKILL.md.
Context: The skill explicitly mandates that commands run under the host's ordinary tool execution policy (approvals/sandbox) and forbids the use of sudo or alternative installers like curl scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw output (stdout and stderr) from external CLI tools and incorporates it into the agent's context and final report, which could be exploited if a tool's output is compromised.
Ingestion points: The trimmed output from version commands and the exact stderr from update failures are recorded and printed back to the agent in SKILL.md.
Boundary markers: The skill uses Markdown tables and fenced code blocks to display output, but does not include explicit instructions for the agent to ignore any command-like text within those sections.
Capability inventory: The skill has the ability to execute shell commands (the update suite) and read/write to the agent's output stream.
Sanitization: No sanitization or filtering is performed; the instructions specify recording "exact stderr" and "trimmed output".

update-all