sentdm__sent-plugin/CONTRIBUTING.md

Contributing

Contributions must be suitable for a public repository and safe to distribute as an agent plugin.

Public-data policy

Never commit:

• credentials, tokens, authorization headers, private keys, or local environment files;
• real customer, contact, message, KYC, billing, or account data;
• private repository links, internal-system names, unpublished operational procedures, or non-public roadmap details;
• local filesystem paths, private network addresses, or production-only diagnostic output;
• generated archives or operating-system metadata.

Use synthetic fixtures with reserved example.com addresses and fictional or reserved phone numbers. If a change is based on non-public material, rewrite it against public documentation and remove source-specific identifiers before opening a pull request.

Development workflow

Edit the canonical files in packages/sent or adapter-sources, then run:

python3 scripts/generate_adapters.py
python3 scripts/validate.py
python3 scripts/test_validation_gates.py
python3 scripts/test_fixtures.py
claude plugin validate . --strict
claude plugin validate ./claude-plugins/sent --strict

Generated changes in root-level plugin.json, mcp.json, skills/, and assets/, plus plugins/sent, claude-plugins/sent, .agents, and .claude-plugin, should be committed with their source changes.

Adding or changing a skill

Treat packages/sent/skills/<skill-name>/ as the source of truth.

1. Keep SKILL.md frontmatter limited to name and description. Write the description for discovery: state what the skill does and include the user intents, product terms, error symptoms, and boundaries that should trigger it.
2. Keep the core workflow concise. Put detailed specifications, examples, and decision tables in skill-local references/; put repeated deterministic checks in scripts/ and test those scripts with synthetic fixtures.
3. Add or update agents/openai.yaml when the skill has OpenAI UI metadata, and keep its display name, short description, and default prompt aligned with SKILL.md.
4. Add positive, negative, and relevant overlap cases in evals/<skill-name>.yaml.
5. Add the skill to the sent dispatcher and to the catalogs in the root README, packages/sent/README.md, and adapter-sources/shared/README.md.
6. Regenerate every adapter and run the complete validation suite above. Do not edit generated skill copies directly.

The repository validator enforces the expected skill set, frontmatter shape, routing coverage, README catalog coverage, local reference integrity, MCP safety contract, generated parity, and public-data policy.