The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's technical mapping documentation recommends the global installation of unvetted third-party utilities from external package registries.
Evidence: references/ops-stack-mapping.md explicitly recommends running npm install -g defuddle to enable markdown extraction functionality.
[COMMAND_EXECUTION]: The operational workflow requires the execution of shell-based commands for system-wide package management and version control tasks.
Evidence: Instructions in references/ops-stack-mapping.md for global npm installs and references/methodology.md for git operations including cloning, committing, and pushing to remote repositories.
[PROMPT_INJECTION]: The skill architecture is highly susceptible to indirect prompt injection due to its core function of aggregating and processing various external, untrusted data sources.
Ingestion points: The skill is instructed to read and process client-provided materials, pitch decks, Slack notes, and website content as defined in the 'INIT' phase of references/methodology.md.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instruction' markers when processing external materials, increasing the risk that instructions hidden in client data could override the agent's behavior.
Capability inventory: The agent is granted capabilities for local file system operations (~/marketing-plans/), shell command execution, and extensive network access via multiple MCP integrations (GitHub, Ahrefs, Stripe, etc.) as detailed in references/ops-stack-mapping.md.
Sanitization: No mention of content sanitization, validation, or escaping of ingested external data is present in the skill's instructions.

marketing-plan