The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing external data from git repositories and pull requests. While this ingestion of untrusted content could potentially harbor malicious instructions (indirect prompt injection), the skill specifically instructs the agent to perform hunk-by-hunk resolution, avoid running code from untrusted forks, and escalate to human review for any ambiguous or suspicious conflicts. These practices effectively mitigate the risk of the agent following adversarial instructions hidden in code conflicts. Evidence Chain: (1) Ingestion points: git merge/rebase (SKILL.md); (2) Boundary markers: human escalation rules (SKILL.md); (3) Capability inventory: git, gh, npm, pnpm, yarn, cargo calls (playbook.md); (4) Sanitization: verification via builds and tests (SKILL.md).
[COMMAND_EXECUTION]: The skill utilizes common development CLI tools such as git, gh, npm, pnpm, yarn, and cargo. These tools are employed for their intended purposes: managing source control, interacting with pull requests, and regenerating dependency lockfiles. The commands provided are standard industry practices for maintaining repository health.
[PERSISTENCE_MECHANISMS]: The skill recommends modifying the global git configuration (git config --global) to enable zdiff3 and rerere. While these changes persist across sessions, they are well-known configuration enhancements specifically designed to aid in the task of conflict resolution and do not represent a security threat.

resolve-merge-conflicts