Record the actual source scope
Distinguish finding a source, reading part of it and studying the requested content. Preserve gaps that affect a conclusion or an unfinished requirement.
XVII.Release 0.17 · source analysis
Actual reading / supported claims / accurate completion
STOLZ A.I. v0.17.0
The optional stolz-evidence skill keeps source analysis tied to material actually inspected. It distinguishes observations, attributed reports, inferences and proposals, and preserves missing requirements through corrections and handoffs.
- Canonical artifact
- 206-file npm package · seven skills
- Verification
- 9 skill cases + 3 short-rule cases
- Update and rollback
- Verified from v0.16.0
- Package smoke
- Windows and Linux
§ 1stolz-evidence
Connect the claim to what was read.
Show what supports the finding
Connect inspected content to the conclusion and its use. Reuse existing context and result identities; a coverage record indexes evidence rather than proving understanding.
Correct and carry the task forward
Correct overstated claims and affected conclusions. Match completion wording to observed actions, and preserve open dependencies in the handoff.
§ 2verification
One bounded local evaluation.
An independent local evaluation covered nine source, claim and handoff cases plus three cases supplied only the short project rule. Inspected answers met the defined checks. Package update, recovery and rollback were checked from the actual v0.16.0 archive.
A procedure with explicit limits
Profiles still install five optimization skills; guard and evidence are installed separately. A dry-run-first helper can merge a short project rule into AGENTS.md, CLAUDE.md or QWEN.md. File merging does not verify native discovery or adherence; the workflow does not guarantee truthfulness, activation or token savings.