The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read, analyze, and resolve code in files with merge conflicts, which may contain adversarial instructions if the changes originate from untrusted sources or external contributors.
Ingestion points: Files containing conflicts are accessed using the Read tool and cat command in SKILL.md.
Boundary markers: Standard git conflict markers (<<<<<<<, =======, >>>>>>>) act as delimiters between different versions of the code, but do not prevent the agent from following instructions embedded within those blocks.
Capability inventory: The agent has access to Bash for command execution and Edit for file modification, providing a surface for carrying out injected instructions.
Sanitization: The skill relies on functional verification via pnpm test and npx tsc, but does not implement specific sanitization or filtering for prompt injection within the code being merged.
[COMMAND_EXECUTION]: The skill uses Bash to perform git operations (git rebase, git merge, git checkout), run build checks (npx tsc), and execute tests (pnpm test). These actions are standard for the task but represent a capability that could be targeted if indirect injection occurs.

conflict-resolution