The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a local shell script scripts/log.sh to append entries to the decision log. The script follows best practices by using set -euo pipefail and avoids dangerous execution patterns like eval.
[DATA_EXFILTRATION]: To verify the log's accuracy, the skill instructs the agent to read session transcripts from platform-specific paths, such as ~/.claude/projects/ or ~/.pi/agent/sessions/. This involves reading metadata and history from the user's home directory, though the instructions explicitly warn the agent to avoid globbing across unrelated projects to protect private data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (e.g., PR titles, filenames, and generated text) which is recorded as 'evidence' in the log file.
Ingestion points: External evidence (PRs, commits) and platform-specific transcript logs are ingested into the agent context.
Boundary markers: The skill uses a TSV (Tab-Separated Values) format and the helper script strips newlines and tabs from inputs to maintain record integrity.
Capability inventory: The skill has file-read access (transcripts) and file-write access (the TSV log).
Sanitization: The scripts/log.sh script includes a clean function that specifically mitigates CSV/Spreadsheet Injection by prefixing characters like =, +, -, and @ with a single quote, preventing malicious payloads from executing if the log is opened in a spreadsheet application.

show-me-your-work