The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads and processes content from local project files that could be manipulated to influence the agent's execution path.
Ingestion points: Reads .konteksto/build-plan.md and .konteksto/progress-tracker.md to identify tasks and determine the execution sequence.
Boundary markers: No delimiters or safety markers are used to isolate the data ingested from these files.
Capability inventory: The skill utilizes Bash, Write, Edit, and Agent tools, which provide a high level of control over the local environment and project files.
Sanitization: While the instructions suggest rejecting unknown selectors, the skill lacks technical sanitization or schema validation for data interpolated into command strings.
[COMMAND_EXECUTION]: The skill dynamically constructs and executes commands using the Bash and Agent tools based on identifiers resolved from project files.
Evidence: The instructions define patterns such as /dev-develop <task ID>, /dev-check verify <task ID>, and /dev-debug <task ID>. If the source files (e.g., build-plan.md) are compromised or contain malicious characters, these identifiers could be used to execute unintended shell commands.

dev-loop