The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads task briefs and findings from a project ledger file (suggested-tasks.decisions.tsv) and interpolates this content into prompts for a 'suggest tool' during the 'Card pass'. Malicious instructions within the ledger or findings could potentially influence agent behavior.\n
Ingestion points: Data is read from <plans>/suggested-tasks.decisions.tsv and captured through the Offer mode's <finding> argument.\n
Boundary markers: The instructions lack specific guidance on using delimiters or safety warnings when passing task briefs to the suggest tool.\n
Capability inventory: The skill executes shell commands via node .agents/pstack/decisions-check.mjs and modifies local files using an append command.\n
Sanitization: There is no evidence of sanitization, validation, or escaping of the content read from the ledger before it is used in subsequent prompts.\n- [COMMAND_EXECUTION]: The skill operates by executing a local script node .agents/pstack/decisions-check.mjs with various subcommands. This script manages the project's task state and interacts with the file system. While the script is part of the project infrastructure, its use constitutes significant command execution capability.

chips