The Agent Skills Directory

[DATA_EXFILTRATION]: The skill handles sensitive Salesforce authentication data. The script scripts/invoke-decision-table.sh executes sf org display to retrieve an accessToken. While the skill documentation correctly warns the agent against leaking or storing these tokens, the automated retrieval and use of credentials in network requests via curl represents an inherent data exposure risk within the agent's execution context.
[COMMAND_EXECUTION]: The helper script scripts/inspect-decision-table-inputs.sh contains a SOQL injection vulnerability. The --decision-table-id argument is interpolated directly into a query string without escaping or validation. An attacker providing a malicious ID could potentially expand the query scope to expose metadata from other records within the DecisionTableParameter object.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as pricing context and product details, which it uses to build API invocation payloads.
Ingestion points: External pricing and quote context are ingested during the workflow outlined in SKILL.md and passed to the helper scripts.
Boundary markers: The instructions lack delimiters or explicit 'ignore' directives to prevent the agent from being influenced by instructions embedded in the external pricing data.
Capability inventory: The skill possesses capabilities including database read access (sf data query), credential access (sf org display), and network operations (curl).
Sanitization: Sanitization is inconsistent across the skill; while find-decision-table.sh uses a basic escaping function for SOQL literals, the injection point in inspect-decision-table-inputs.sh is unmitigated.

revenue-cloud-decision-table