ADR-0168 - Render structured chat output

Context

Dissect and Bulletproof wrapped Markdown reports in text fences. Tables became monospace blobs; labels lost hierarchy; evidence became harder to scan. The structure was useful. The wrapper was not.

Suspec has two output classes: durable or transient artifacts written by sus-* authors, and direct chat from universal methods. Chat is a rendered interface, not a file attachment pretending to be one.

Decision

1. Artifact authorship stays closed. Only sus-* artifact authors create Suspec artifacts. Structured output alone does not justify a file.
2. Chat structure renders as Markdown. Use headings, bold labels, prose, lists, and normal tables. Never wrap a report, table, or prose handoff in a fence.
3. Fences carry literal material only. Code and untouched raw evidence may use an appropriate fence. Artifact templates may show fenced source because their destination is a file, not chat.
4. A summary may follow a report. Keep it brief and add it only when it improves orientation. Never restate the report.
5. The catalog gate owns the rule. Every skill Output section rejects fenced content. Method- specific gates preserve required labels, tables, evidence, and unknown-state handling.

Narrowed decisions

• ADR-0109: economical output remains; readability is not waste.
• ADR-0157: universal methods still return chat and artifact authorship remains closed.
• ADR-0161: constrained handoffs use rendered chat structure unless they are literal code, raw evidence, or artifact templates.

Consequences

• Structured reports remain compact and become scannable.
• A formatting requirement cannot accidentally create an artifact.
• Literal code and proof retain exact formatting without swallowing the surrounding report.

Status

Accepted (2026-07-22). Narrows ADR-0109, ADR-0157, and ADR-0161.