The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and process multiple markdown and JSON artifacts (e.g., 10-intent.md, 20-checkpoint.md, 90-evidence.md) from the project's documentation folder (docs/aegis/work/) and CONTEXT.md. This ingestion creates a vulnerability where instructions embedded in these files by an external actor could influence the agent's behavior during task resumption.
Ingestion points: Accesses and interprets various project-level files including docs/aegis/work/YYYY-MM-DD-<slug>/ artifacts and CONTEXT.md to establish task state.
Boundary markers: The instructions do not define delimiters or specific safety warnings to differentiate data from control instructions within these artifacts.
Capability inventory: The skill possesses the ability to execute shell commands via a workspace helper script and perform file system operations.
Sanitization: No data validation or sanitization routines are specified for the content processed from the work records.
[COMMAND_EXECUTION]: The protocol utilizes a Python-based helper tool (aegis-workspace.py) to automate lifecycle records and structure the workspace.
Evidence: The skill directs the agent to execute shell commands such as python <aegis-workspace-helper> init, new-work, add-checkpoint, and bundle to maintain project state.

long-task-continuation