The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data to generate project records, creating a surface for indirect prompt injection.
Ingestion points: The agent is instructed to read DEVLOG.md, ADR files, and git commit subjects to synthesize updates for the CHANGELOG.md and other records (identified in SKILL.md and scripts/project-records.ps1).
Boundary markers: The skill relies on standard Markdown list structures for generated output. It includes explicit instructions to redact secrets and to show all drafts to the user for confirmation before committing (as seen in SKILL.md and references/devlog.md).
Capability inventory: The skill can write to the local filesystem and execute git commands via the provided scripts/project-records.ps1 script.
Sanitization: The automation script uses regular expressions for structural validation and implements a custom argument escaping function (ConvertTo-RecordsArgument) to sanitize data passed to shell commands.
[COMMAND_EXECUTION]: The skill relies on a custom PowerShell script, scripts/project-records.ps1, to interact with the repository's git history and configuration.
The script executes various git subcommands (e.g., git log, git diff, git rev-parse) to automate record validation and rendering.
To mitigate command injection risks, the script uses a dedicated escaping utility, ConvertTo-RecordsArgument, for all process invocations.
The skill instructions emphasize using pathspecs for all commits to prevent accidental inclusion of untracked or unwanted files, and advise against automated pushing of changes.

tl-project-records