/workspace-tests — cargo test across N Rust workspaces

Walks a configured list of in-scope Rust workspaces and runs cargo test --workspace against each. Fails fast (stops at first red workspace) by default; --continue-on-error runs the full set. Reports total + per-repo pass/fail counts.

Why this exists

Surfaced 2026-05-20 from PlausiDen-Forge #303: an opportunistic hygiene cycle that ran cargo test --workspace only on the currently-active repo silently missed a real E0063 in Loom (CmsSection::Form scaffold missing style field). Cross-repo discovery only happened when the next agent ran cargo fmt and hit the compile error.

Real failure mode: a substrate-wide field addition lands in one crate but the consumer-side scaffolds in sibling repos don't update. Tests still pass in the producer crate; consumer crate breaks at HEAD. Without explicit cross-repo coverage, hours of "green builds" accumulate before the gap surfaces.

Steps

1. Read the workspace list

Default in-scope set (PlausiDen substrate):

/home/paul/projects/PlausiDen-Forge
/home/paul/projects/PlausiDen-Loom
/home/paul/projects/PlausiDen-Crawler
/home/paul/projects/Crucible

LFI repos (PlausiDen-LFI, Forge-LFI) are excluded by default — they have a dedicated Claude instance per memory feedback_lfi_out_of_scope_for_this_instance. Add --include-lfi to override.

User can override the set via:

• $WORKSPACE_TESTS_REPOS env (colon-separated absolute paths)
• ~/.config/workspace-tests/repos.txt (one path per line)

2. Per repo, run

sudo -u paul cargo test --workspace --manifest-path "$REPO/Cargo.toml" 2>&1 \
  | tee "/tmp/workspace-tests-$(basename "$REPO").log" \
  | grep -E '^test result|FAILED|^error'

Capture the trailing aggregate counts via:

sudo -u paul cargo test --workspace --manifest-path "$REPO/Cargo.toml" 2>&1 \
  | grep '^test result' \
  | awk '{s+=$4; f+=$6} END {print s,"passed,",f,"failed"}'

Skip repos with no Cargo.toml at the root (surface a warning; don't fail).

3. Fail-fast vs continue

• Default: bail at first non-zero exit. Report which repo broke.
• --continue-on-error: run all repos, then exit non-zero if any failed.

4. Output format

PlausiDen-Forge       1881 passed, 0 failed
PlausiDen-Loom        1335 passed, 0 failed
PlausiDen-Crawler     1084 passed, 0 failed
Crucible                15 passed, 0 failed
-----
TOTAL                 4315 passed, 0 failed across 4 repos

On failure, the broken repo's tail-50 of test output prints verbatim before the summary, with a === <repo>: REGRESSION === banner so the agent / operator can grep for the boundary.

5. Bypass conditions

• Empty workspace list → exit 0 with "no repos configured" warning
• cargo not installed → exit 2 with the install URL
• Any repo path doesn't exist → warn + skip, don't fail

6. Recommended invocation

Inline in /loop hygiene priorities:

sh ~/.local/share/claude-tools/lib/workspace-tests.sh --continue-on-error

Or, when the loop fires and you have ~30s of budget:

sh ~/.local/share/claude-tools/lib/workspace-tests.sh
# fails fast — first red repo aborts; agent then drives into the fix

Related skills

• regression-guard — adjacent: catches per-PR regressions on a single repo
• loop-health — checks loop / cron state but not workspace test state
• audit-unused-dep — per-repo dep audit; complements workspace-tests as a hygiene pair

Don't

• Don't recurse into nested Cargo.tomls (only the workspace root).
• Don't run cargo build separately — cargo test --workspace exercises the full compile path.
• Don't auto-fix detected failures. The skill is read-only; the agent decides whether to attempt a fix in the surfacing repo.
• Don't include LFI repos by default per the dedicated-instance doctrine.