The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a process for recording research ideas, predictions, and hindsight which are stored in a markdown file and later retrieved during a 'Monthly Review'. This pattern creates a surface where data provided by a user (or external sources) could influence agent behavior when read back into the context.
Ingestion points: User input for titles and fields are passed to the append function in scripts/journal.py and written to the journal file.
Boundary markers: Entries are delimited by dated headers (e.g., ## YYYY-MM-DD kind: title). However, the script does not escape or sanitize field values, which could allow the introduction of markdown structures or agent instructions within the file.
Capability inventory: The skill possesses file read/write capabilities via scripts/journal.py and the ability to execute the journal script through a test runner in scripts/test_journal.py.
Sanitization: The script validates that titles do not contain newlines, but it does not perform similar checks or escaping on the values passed to the --field arguments, allowing for multi-line content to be injected into the journal file.
[COMMAND_EXECUTION]: The scripts/test_journal.py file utilizes subprocess.run to execute the bundled journal.py script. This is performed within a test context to verify append-only behavior, but it represents the spawning of external processes from within the skill's script directory.

credo-journal