Skip to content

Artifact gallery

Every Anatomize output has one owner and one lifecycle purpose. Canonical JSON is the interchange contract; Markdown and text are projections from the same validated object.

Artifact Produced by Use it for Usually review as
Capability document review capabilities Negotiate operations, profiles, target kinds, schemas, and interaction rules JSON
Review session review start Reuse one exact evidence state across questions JSON
Dossier exchange review dossier / expand Answer one bounded orientation, design, audit, localisation, implementation, change-review, or closure question Markdown plus JSON when traceability matters
Similarity artifact review similarity Discover conservative implementation, test, and documentation candidates JSON
Consolidation dossier review consolidate Decide whether one possible duplicate should be merged, kept, or investigated further Markdown
Decision record (overlay) review overlay-create Preserve your disposition, owner, rationale, and review conditions separately from observed evidence JSON or Markdown
Overlay evaluation review overlay-check Determine whether relevant evidence changed after a decision Markdown
Implementation intent review intent Bind explicit obligations to the exact before-state JSON
Change dossier review change Compare canonical evidence across independent states Markdown plus JSON
Closure report review verify Account for every obligation against fresh evidence Markdown plus JSON
Artifact check review check Validate schema, digest, identity, state, and references without mutation JSON

Reading a dossier

Read in this order:

  1. Status — whether the evidence is complete, partial, or blocked.
  2. Question and targets — the exact decision boundary the dossier answers.
  3. Evidence groups — why each fact matters, not merely where it was found.
  4. Conflicts — incompatible source-bound claims preserved without ranking.
  5. Omissions and limitations — absent evidence and what the methods cannot establish.
  6. Expansion actions — exact, bounded ways to obtain more evidence.

The selection reasons are part of the output contract. They make a compact context human-reviewable and help an agent distinguish essential evidence from optional orientation.

Choosing an output format

Use --format json between processes and independent releases. JSON artifacts have strict current schemas, deterministic ordering and serialization, stable identity, and explicit version negotiation.

Use --format markdown for durable code-review attachments and design records. Headings and labels preserve conflicts, omissions, limitations, and recovery actions without relying on colour.

Use --format text for terminals and logs. It is concise and ANSI-free, but it is not an interchange format and consumers should not parse it.

Source and privacy

A session excludes repository source text by default. Exact --include-source paths make selected content portable for a trusted consumer, but also change the publication boundary. Imported provider text is untrusted and can contain private material even when Anatomize validates its structure.

Before publishing an artifact:

  • prefer a Markdown projection over a complete machine session;
  • inspect explicitly included source and provider content;
  • retain repository-relative paths and content digests;
  • do not treat artifact encoding as secret scanning; and
  • publish only the evidence needed to understand the decision.

Schemas and compatibility

The generated schema catalogue provides reviewable and downloadable JSON Schemas. Negotiate current schema versions using review capabilities. There is no migration layer: regenerate an older or unknown artifact with the current producer rather than editing its version, identity, or digest.