The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local audit reports to determine its execution flow and content generation.
Ingestion points: The skill reads reports/hubspot-audit-*.md to extract grades, metrics, and recommendations.
Capability inventory: The skill is capable of writing new files to the filesystem (implementation plans and new SKILL.md files) and performing Git operations (git commit, git push, gh pr create).
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore or sanitize instructions that might be embedded within the audit reports.
Sanitization: There is no evidence of input validation or escaping for the data extracted from the reports before it is used to generate the plan or new skills.
[DYNAMIC_EXECUTION]: The skill features a mechanism to automatically generate new AI agent skills when it detects gaps in its current capabilities.
Evidence: Step 6 ("Handle Gaps — Missing Skills") describes a process where the agent creates new SKILL.md files following a standard format based on findings in the audit.
Risk: If the source audit report is poisoned with malicious recommendations, the agent may generate new skills with dangerous instructions or behaviors.
[COMMAND_EXECUTION]: The skill utilizes command-line tools to manage repository contributions.
Evidence: The skill uses git and the GitHub CLI (gh) to fork repositories, create branches, push code, and open pull requests.
Context: These operations target the author's own repository (tomgranot/hubspot-admin-skills), which is considered a legitimate vendor resource in this context.

hubspot-implementation-plan