The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests agent transcripts to identify user preferences. These transcripts are untrusted sources that could contain malicious instructions designed to influence the drafted skill.
Ingestion points: Transcript files are located and read in SKILL.md (Step 1) using the find-transcripts tool.
Boundary markers: Instructions do not specify delimiters or warnings for the subagents to ignore instructions embedded within the mined text.
Capability inventory: The skill has file-write capabilities (Step 4), git command execution (Step 0, Step 6), and interactive user prompts.
Sanitization: No content sanitization or validation of the mined transcript data is specified.
[COMMAND_EXECUTION]: The skill utilizes shell commands to interact with version control systems.
Evidence: SKILL.md (Step 0 and Step 6) specifies the use of git log, git worktree, git commit, and pull request operations to manage the skill files.

automate-me