The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository's changelog files and project structure to perform classification and updates. However, it implements several safeguards:
Ingestion points: Processes CHANGELOG.md, DEVELOPER_CHANGELOG.md, and repository files (e.g., package.json, app.json) via bundled helpers scripts/setup.ts and scripts/query.ts.
Boundary markers: Utilizes standard Markdown headings (e.g., ## Unreleased, ## 1.0.0) and specific HTML comments () as structured delimiters.
Capability inventory: Accesses file-write operations to maintain changelogs and policy files; spawns git, bun, and awk processes via local scripts for repository inspection and synchronization.
Sanitization: Implements a deterministic Markdown parser in scripts/lib/changelog-parse.ts and enforces strict schema validation using multiple .schema.json files for configuration, requests, and receipts.
[COMMAND_EXECUTION]: The skill utilizes local TypeScript and shell scripts (scripts/setup.ts, scripts/query.ts, scripts/check-fork-sync.sh) to automate repository management. These scripts are executed via the bun runtime and interact with git to perform status checks, file comparisons, and version identification. All commands are scoped to the repository's management and do not exhibit suspicious behavior such as privilege escalation or network exfiltration.
[DATA_EXPOSURE]: The skill accesses local repository configuration files (e.g., .simple-changelogs.json) and project manifests. There is no evidence of the skill attempting to access sensitive user directories, environment variables, or credentials outside the repository's scope.

simple-changelogs-web