The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local project files to generate recommendations, which can be exploited if those files contain malicious instructions.
Ingestion points: The skill uses cat and grep to read files such as package.json, pyproject.toml, and Cargo.toml in SKILL.md.
Boundary markers: There are no explicit instructions or delimiters to isolate or ignore instructions embedded within the ingested codebase files.
Capability inventory: The skill utilizes ls, cat, and grep. While the skill's primary output is markdown recommendations, it suggests the creation of other automations with write_to_file and run_command capabilities.
Sanitization: No sanitization or validation logic is defined for the content extracted from the codebase files.
[DATA_EXFILTRATION]: The skill scans for sensitive file paths to inform its security automation recommendations.
Evidence: Phase 1 in SKILL.md and patterns in references/hooks-patterns.md check for the existence of .env, .env.local, credentials.json, and secrets.yaml using ls and presence of checks.
Context: This identification is used to recommend 'Protection Hooks' intended to prevent accidental exposure or modification of these sensitive files.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation and use of various external tools and services.
Evidence: Recommends installing Model Context Protocol (MCP) servers such as context7, Supabase, Convex, and Sentry, as well as plugins like anthropic-agent-skills.
Context: These recommendations target well-known services and industry-standard tools for extending agent capabilities.
[DYNAMIC_CONTEXT_INJECTION]: The documentation provides instructions and examples for using the platform's dynamic context injection feature.
Evidence: references/skills-reference.md explains and provides examples of the !command syntax, such as !git branch --show-current and !gh pr diff.
Context: This content is educational reference material designed to help users build their own custom skills using available platform features.

claude-automation-recommender