The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external GitHub issues and pull requests and writes it directly into files that govern the agent's behavior.
Ingestion points: Data is retrieved via gh issue list and gh pr list commands within the main function of scripts/convention.mjs.
Boundary markers: The generated content is enclosed in  and  markers. While these delineate the block, they do not inherently prevent an LLM from following instructions embedded within that block.
Capability inventory: The script uses writeFileSync to modify CLAUDE.local.md or AGENTS.local.md and updates the .gitignore file.
Sanitization: The script extracts specific metadata (prefixes, headings) but does not escape or sanitize the body text of issues and PRs before writing them to the instruction files.
[COMMAND_EXECUTION]: The skill relies on executing local system utilities to function.
It calls git rev-parse --show-toplevel to locate the repository root.
It uses the GitHub CLI (gh) to fetch repository metadata, issue lists, and pull request history. All commands are executed via spawnSync using array-based arguments, which mitigates shell injection risks.

convention