The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to update the thread log using a shell append command rather than standard file system APIs.
Evidence: In references/instructions/append-log-line.md, the instruction printf '%s\n' '- (type) gist with the reason folded in' >> <thread root>/log.md is provided for updating the thread log.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest multiple project-level documents which could contain untrusted instructions that influence agent behavior during discussions.
Ingestion points: The skill reads AGENTS.md, docs/glossary.md, decision records in docs/adr/ and docs/pdr/, seed.md, log.md, spec.md, and the delta/ directory as specified in SKILL.md under ## Inputs.
Boundary markers: No explicit delimiters or boundary instructions are defined to separate untrusted document content from the agent's internal instructions.
Capability inventory: The skill has the ability to append to files on disk using shell-based printf commands.
Sanitization: There are no explicit instructions to sanitize or escape the content of the 'gist' or 'reason' before they are interpolated into the shell command template.
[DYNAMIC_EXECUTION]: The skill generates and executes shell commands dynamically based on the outcome of the user interview, which may involve untrusted input.
Evidence: The logging mechanism in references/instructions/append-log-line.md constructs a shell command string at runtime using variables derived from conversation context without enforced sanitization.

discussion