The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data ingested from external sources (Notion pages from steps 2-11 of the GTM flow) and uses this data to generate recommendations. This creates a potential surface for indirect prompt injection if those external pages contain adversarial content.
Ingestion points: Data is read from existing Notion pages identified as steps 2 through 11 of the GTM process.
Boundary markers: The prompt does not specify the use of strict delimiters (e.g., XML tags or block quotes) when interpolating the retrieved data into the model context.
Capability inventory: The skill has the capability to write content back to Notion (creating a new sub-page '12 · Docs Plan').
Sanitization: No explicit sanitization or validation of the input data from Notion is described in the instructions.
[SAFE]: The skill references a local file at ${CLAUDE_PLUGIN_ROOT}/reference/agency-stages.md. Accessing local reference documentation included with the skill package is a standard and safe operation.
[SAFE]: The skill uses Ukrainian language for its output and logic, which is consistent with its stated purpose of writing to Notion in Ukrainian. This is legitimate localized content and does not constitute obfuscation.

gtm-docs-plan