The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the !command`` syntax to execute a shell command (cat .agents/product-marketing-context.md) at load time. This is used to incorporate local context files into the agent's environment, which executes silently when the skill is accessed.
[PROMPT_INJECTION]: User-provided arguments ($ARGUMENTS) are directly interpolated into the system prompt instructions ('focus on: $ARGUMENTS'). This creates a surface where a user could potentially supply instructions that attempt to override the agent's intended behavior.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from multiple external sources including Stripe, analytics databases, and manual file uploads (CSV/JSON).
Ingestion points: Customer data retrieved via MCP tools from Stripe (emails, payment feedback), product activity from databases, and manual CSV/TSV/JSON uploads.
Boundary markers: No specific delimiters or safety instructions are used to wrap the ingested data to prevent it from being interpreted as instructions by the model.
Capability inventory: The skill can execute a local Python utility script (ds_utils.py) via the Bash tool and perform automated data retrieval through various MCP tools.
Sanitization: The skill does not perform explicit sanitization or filtering of the content retrieved from external data sources before processing it through the LLM.
[COMMAND_EXECUTION]: The skill relies on the execution of a Python script located outside the skill's own directory (../../scripts/ds_utils.py). The Bash tool is restricted to this specific script pattern, which the skill uses for deterministic calculations such as MRR and churn rate analysis.

ds-churn-signals