The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from context modules (.md files) and evaluation goldensets within the repository.
Ingestion points: Scans files in context_modules/ (rules, personas, examples) and evals/ directories, as well as project configuration in .adlc/init-options.json.
Boundary markers: The skill relies on YAML frontmatter delimiters (---) and specific markdown headers to identify metadata, though the body content is processed for index generation.
Capability inventory: Includes file reading/writing, executing shell commands (find, git log, python3), and making LLM calls to evaluate directive necessity in Phase 10.
Sanitization: Uses JSON parsing via Python for configuration data and performs string truncation for descriptions in the CDR.md index.
[DYNAMIC_EXECUTION]: The skill performs runtime script generation and execution for file management.
Evidence: Utilizes python3 - and cat <<EOF patterns in team-helpers.sh to dynamically assemble and inject managed instruction sections into AGENTS.md and rebuild manifest files like .skills.json.
Context: These operations are restricted to project-local files and follow static templates defined within the skill's helper scripts.
[COMMAND_EXECUTION]: The skill executes standard system utilities to maintain the repository state.
Evidence: Invokes find for file discovery, git log to generate audit trails in log.md, and python3 for robust JSON parsing and search-and-replace operations.
Context: Commands are used strictly for repository indexing and do not involve network exfiltration or sensitive path access.

team-repair