Thanks for taking the time to contribute.
This repo is a public OSS extraction of the SROS Self-Compiler prompt-spec. The core rule is: keep it clean, chat-friendly, and receipts-first.
- Output formats: JSON and YAML modes that preserve the same semantics as XML
- Validator tooling:
- single-root output checks
- URI sanitation checks
- numeric score bans
- Better examples and docs
- Stronger governance templates and approval contracts
- Adapter registry expansions (still abstract, no provider coupling in OSS v1)
- Claims of cryptographic determinism
- Provider-specific model routing baked into receipts
- Any automation that requires credential harvesting or private data extraction
- Open an issue with:
- Problem statement
- Proposed behavior
- Minimal example input and expected output
- If accepted, open a PR with:
- The patch
- Updated example(s) if behavior changed
- Updated README if user-facing
- Do not introduce markdown links inside XML fields.
- Do not introduce Google-wrapped URIs or link wrappers.
- Keep exactly one root output per response.
- Do not add numeric trust scores or numeric lens scores.
- Use
score_class="high|medium|low"if you need classification.
- Hash fields may remain placeholders:
status="optional" kind="placeholder"
- If you implement hashing, include:
- canonical byte serialization definition
- deterministic test cases
- a validator script and fixtures
At minimum, add or update fixtures that cover:
- Fast mode
compile:returnspromptunit_package - Vague input returns
compiler_input_form - XML spec paste triggers intake mode
- URI sanitation blocks link-wrapped URIs
- Numeric scoring strings do not appear in output
- No markdown links in XML fields
- No
google.com/search?q=anywhere - No numeric scoring fields
- Examples updated if behavior changed
- README updated if user-facing change
If you want attribution, add yourself to a CREDITS.md section:
- Name
- Handle
- Contribution type