The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands to assess and resolve conflicts, including git status, git checkout, git add, and git commit. It also invokes custom shell scripts located at .forge/skills/resolve-conflicts/scripts/handle-deleted-modified.sh and .forge/skills/resolve-conflicts/scripts/validate-conflicts.sh.
[DYNAMIC_EXECUTION]: To resolve generated files, the skill executes project-defined build and generation commands such as npm run build, make generate, and graphql-codegen. These commands run code defined within the repository being merged, which could be untrusted.
[EXTERNAL_DOWNLOADS]: The skill uses package managers (npm, cargo, yarn, poetry, bundle) to regenerate lock files. This process involves fetching dependencies from external registries.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes external, untrusted data from conflicted files.
Ingestion points: The agent reads file content and conflict markers during the assessment and resolution phases (SKILL.md Step 1 and Step 4).
Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between its own logic and the content within the conflicted files.
Capability inventory: The agent has the ability to execute shell commands, build projects, and commit changes to the repository.
Sanitization: There is no evidence of sanitization or filtering of the content read from conflicted files before it is processed.

resolve-conflicts