The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted sources within the repository to guide its operations, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill reads STATUS.md, AGENTS.md, README.md, handoff documents, and issue trackers from the repository to identify project-specific installation commands and determine the next items of work.
Boundary markers: No explicit delimiters or boundary instructions are used to isolate the content of these repository files from the agent's instructional context.
Capability inventory: The skill possesses the capability to execute shell commands via git, gh, and bash, as well as running arbitrary repository-defined installation commands.
Sanitization: There is no evidence of validation or sanitization of the commands or task descriptions retrieved from the project files before they are acted upon.
[COMMAND_EXECUTION]: The skill performs a variety of shell operations to automate repository maintenance.
It executes internal Bash scripts provided with the skill (classify.sh, delete_head.sh, wait_merged.sh) to perform branch analysis and PR monitoring.
It runs repository-specific installation commands (e.g., npm install, pnpm install, pip install) when it detects changes to lockfiles.
For Adobe-related repositories, it routes Git and GitHub CLI operations through a local environment script at ~/Projects/adobe/scripts/with-adobe-route.
[DATA_EXPOSURE]: The classify.sh script accesses the agent's internal state database at ~/.t3/userdata/statev2.sqlite to identify worktrees currently bound to active agent threads, preventing the deletion of worktrees that are still in use.

merged