Use these patterns to generate, query, and publish Compass artifacts in CI without depending on human text or hiding failures.

Cookbook: CI and automation

Use these patterns to generate, query, and publish Compass artifacts in CI without depending on human text or hiding failures.

Recipe 1: build and upload a structural graph

set -eu
compass --version
compass update . --no-viz
test -s compass-out/graph.json
test -s compass-out/GRAPH_REPORT.md
test -s compass-out/manifest.json

Upload the three artifacts plus:

git rev-parse HEAD > compass-out/source-revision.txt
compass --version > compass-out/compass-version.txt

Treat graph artifacts with source-code confidentiality.

Recipe 2: exact policy query

Create .compass/no-ambiguous-auth.cypher:

MATCH (source)-[edge]->(target)
WHERE edge.confidence = 'AMBIGUOUS'
  AND (source.label CONTAINS 'Auth' OR target.label CONTAINS 'Auth')
RETURN source.id, type(edge), target.id
ORDER BY source.id, target.id
LIMIT 500

Run:

compass query --cql \
  --file .compass/no-ambiguous-auth.cypher \
  --format json \
  --output compass-out/no-ambiguous-auth.json

Decide policy in a separate script that validates compass.cql.result/1 and counts rows. This keeps query execution and organization-specific pass/fail rules separate.

Recipe 3: parameterized reusable query

.compass/callers.cypher:

MATCH (caller)-[edge:CALLS]->(target)
WHERE target.label = $target
RETURN caller.id, edge.confidence, target.id
ORDER BY caller.id
LIMIT 500

Run:

compass query --cql \
  --file .compass/callers.cypher \
  --param target="$TARGET_LABEL" \
  --format json \
  --output compass-out/callers.json

Do not inject TARGET_LABEL into source text.

Recipe 4: compare base and head

set -eu
compass history build "$BASE_SHA" --code-only
compass history build "$HEAD_SHA" --profile-from "$BASE_SHA"
compass diff "$BASE_SHA" "$HEAD_SHA" \
  --format json > compass-out/semantic-diff.json

Use full commit SHAs supplied by the CI provider. Avoid fetch-on-demand inside historical materialization; make required commits available during checkout.

Cache strategy

Cache keys should include:

operating-system / target
Compass version or binary hash
Cargo.lock or release artifact identity
build profile / code-only vs semantic
relevant parser/extractor version
repository content key

Do not restore:

a semantic cache under another model/prompt/profile;
query plan cache under an incompatible graph schema;
a manifest from another output/root;
a live SQLite/WAL history copy assembled from partial files.

History is a live durable store. Prefer supported backup or rebuild workflows over naive cache archiving.

Failure policy

Compass nonzero
  -> job fails; retain diagnostic

Query limit
  -> job fails or reports inconclusive; never "zero matches"

Missing provider key
  -> fail semantic job or explicitly run code-only

Unknown result major version
  -> fail closed

Graph policy rows found
  -> organization policy decides warn/fail

Security

Pin or verify downloaded release artifacts.
Do not echo provider or database keys.
Avoid semantic providers for unapproved repositories.
Do not expose graph.html publicly by default.
Sanitize CI artifacts according to repository classification.
Do not execute untrusted checkout build scripts merely to build a graph.

Suggested artifact bundle

compass-out/
├── graph.json
├── GRAPH_REPORT.md
├── manifest.json
├── source-revision.txt
├── compass-version.txt
├── policy-result.json
└── topology-diff.json

HTML is optional and can be omitted for size or security.

Next step: implement the build-only recipe first, inspect its artifacts, then add one exact policy query with explicit schema validation.

Cookbook: CI and automation

On this page