Summary
A source-level build-artifact hygiene audit of tensor4all/strided-rs main (commit dc0a8e03286c61a84d56446b5cc2c53295f75d76, "build: reduce Cargo artifact size (#228)") against the shared build-artifact controls added in tensor4all/tensor4all-agent-rules#9 found one confirmed, non-duplicate failure: the repository documents no stale-artifact pruning mechanism.
PASS/FAIL table
| Control |
Shared-rule basis |
Result |
Control 1 — Default profiles (dev/test debug = 0, overrides preserved, no full release debuginfo) |
rules/rust/performance.md:51-59, rules/common/docs-and-tests.md:63-66 |
PASS |
Control 2 — Comprehensive CI profile (incremental = false, strip = "symbols"; coverage job exempt) |
rules/rust/performance.md:60-61, rules/common/docs-and-tests.md:55,67-68 |
PASS |
| Control 3 — Explicit local pre-PR attestation for hosted-CI-measured gates |
rules/common/docs-and-tests.md:69-73 |
PASS (see note) |
| Control 4 — Documented stale-artifact pruning mechanism |
rules/rust/performance.md:62-65, rules/common/docs-and-tests.md:74-78 |
FAIL |
Control 1 — PASS
Cargo.toml:54-55 sets [profile.dev] debug = 0; Cargo.toml:57-58 sets [profile.test] debug = 0. debug-assertions, overflow-checks, and incremental keep their Cargo defaults.
- No
[profile.release] override exists, so release uses Cargo's default debug = false (no full debuginfo).
- The one-command debugger override is available and unblocked:
CARGO_PROFILE_DEV_DEBUG=1 / CARGO_PROFILE_TEST_DEBUG=1 (pattern demonstrated at docs/2026-08-09-cargo-artifact-reduction.md:24,46).
Control 2 — PASS
.github/workflows/ci.yml:41 runs hosted tests with cargo test --profile ci (also facade doctests at ci.yml:44).
Cargo.toml:61-64 defines [profile.ci] inherits = "test"; incremental = false; strip = "symbols".
- Coverage job (
.github/workflows/ci.yml:60-73) runs cargo llvm-cov --workspace with its own instrumentation-owned profile and is exempt from symbol stripping, consistent with docs/2026-08-09-cargo-artifact-reduction.md:80-81 and the audit's coverage exemption.
Control 3 — PASS (note)
The draft audit originally flagged this control as failing, but that finding was removed as unsupported. rules/common/docs-and-tests.md:69-73 states the local pre-PR gate may be attestation-based — a permission, not an obligation — while rules/common/docs-and-tests.md:53-55 assigns coverage enforcement to hosted CI. That enforcement exists: .github/workflows/ci.yml:71-73 runs scripts/check-coverage.py against per-file thresholds and fails the job when thresholds are missed. AGENTS.md:53-60 documents the local pre-PR gate (cargo fmt --all -- --check, cargo test --workspace) per the local-validation guidance; the repository has no local coverage gate, and the policy imposes none.
Control 4 — FAIL (confirmed)
rules/rust/performance.md:62-65 requires: "Document the pruning mechanism (an age-based sweep tool, or a periodic cargo clean), and propose a cleanup when target/ growth is dominated by stale artifacts." rules/common/docs-and-tests.md:74-78 likewise states repositories "should document a pruning mechanism (an age-based sweep tool or a periodic full clean)". Current main documents none:
AGENTS.md — no mention of artifact pruning or cargo clean anywhere.
REPOSITORY_RULES.md — no mention of artifact pruning.
docs/2026-08-09-cargo-artifact-reduction.md — documents profile-based size reduction (debug = 0, stripped ci profile) and a rejected integration-test consolidation experiment, but prescribes no pruning mechanism.
docs/superpowers/plans/2026-07-04-fused-elementwise-dag.md:13 — contains a plan-scoped note "run du -s . before work; if over 100GB, propose cleanup before continuing". This is a dated plan instruction for one issue, not a repository-wide documented pruning policy.
.github/workflows/ci.yml:46-50 — CI reports du -sh target and df -h target for visibility but performs no pruning.
- No
check-pr*.sh/cleanup script, Makefile/Justfile, or scheduled workflow addresses stale-artifact removal.
Expected remediation
Document a concrete pruning mechanism in repository-level documentation (AGENTS.md, REPOSITORY_RULES.md, or a dedicated docs/ file). Minimum viable options:
- Prescribe a periodic
cargo clean cadence (e.g., weekly, or after every N dependency bumps), and/or
- Document an age-based sweep command (e.g.,
find target -type f -mtime +14 -delete after confirming no build is in progress) with usage notes, and
- Add a trigger to propose a cleanup when
target/ growth is clearly dominated by stale artifacts no current build uses, optionally integrated as a scheduled CI workflow that prunes the runner cache.
Scope and exclusions
- This audit is source-level only. It did not inspect or estimate local
target/ size and ran no cargo build, cargo test, cargo clean, or du on target/; local target/ size is owned by a separate parent-AGENTS check and is explicitly excluded here.
- Coverage/instrumentation profiles are exempt from symbol stripping and were not audited against the strip requirement.
- Related open PRs: none currently open address this control (
gh pr list --state open returns none). No existing issue covers it (gh issue list --state all searched for artifact/target/pruning/coverage-attestation topics).
Policy authority
tensor4all/tensor4all-agent-rules#9 — PR that added the build-artifact controls.
rules/rust/performance.md:62-65 and rules/common/docs-and-tests.md:74-78 — the stale-artifact pruning documentation requirements.
Summary
A source-level build-artifact hygiene audit of
tensor4all/strided-rsmain(commitdc0a8e03286c61a84d56446b5cc2c53295f75d76, "build: reduce Cargo artifact size (#228)") against the shared build-artifact controls added intensor4all/tensor4all-agent-rules#9found one confirmed, non-duplicate failure: the repository documents no stale-artifact pruning mechanism.PASS/FAIL table
dev/testdebug = 0, overrides preserved, no full release debuginfo)rules/rust/performance.md:51-59,rules/common/docs-and-tests.md:63-66incremental = false,strip = "symbols"; coverage job exempt)rules/rust/performance.md:60-61,rules/common/docs-and-tests.md:55,67-68rules/common/docs-and-tests.md:69-73rules/rust/performance.md:62-65,rules/common/docs-and-tests.md:74-78Control 1 — PASS
Cargo.toml:54-55sets[profile.dev] debug = 0;Cargo.toml:57-58sets[profile.test] debug = 0.debug-assertions,overflow-checks, and incremental keep their Cargo defaults.[profile.release]override exists, so release uses Cargo's defaultdebug = false(no full debuginfo).CARGO_PROFILE_DEV_DEBUG=1/CARGO_PROFILE_TEST_DEBUG=1(pattern demonstrated atdocs/2026-08-09-cargo-artifact-reduction.md:24,46).Control 2 — PASS
.github/workflows/ci.yml:41runs hosted tests withcargo test --profile ci(also facade doctests atci.yml:44).Cargo.toml:61-64defines[profile.ci] inherits = "test"; incremental = false; strip = "symbols"..github/workflows/ci.yml:60-73) runscargo llvm-cov --workspacewith its own instrumentation-owned profile and is exempt from symbol stripping, consistent withdocs/2026-08-09-cargo-artifact-reduction.md:80-81and the audit's coverage exemption.Control 3 — PASS (note)
The draft audit originally flagged this control as failing, but that finding was removed as unsupported.
rules/common/docs-and-tests.md:69-73states the local pre-PR gate may be attestation-based — a permission, not an obligation — whilerules/common/docs-and-tests.md:53-55assigns coverage enforcement to hosted CI. That enforcement exists:.github/workflows/ci.yml:71-73runsscripts/check-coverage.pyagainst per-file thresholds and fails the job when thresholds are missed.AGENTS.md:53-60documents the local pre-PR gate (cargo fmt --all -- --check,cargo test --workspace) per the local-validation guidance; the repository has no local coverage gate, and the policy imposes none.Control 4 — FAIL (confirmed)
rules/rust/performance.md:62-65requires: "Document the pruning mechanism (an age-based sweep tool, or a periodiccargo clean), and propose a cleanup whentarget/growth is dominated by stale artifacts."rules/common/docs-and-tests.md:74-78likewise states repositories "should document a pruning mechanism (an age-based sweep tool or a periodic full clean)". Currentmaindocuments none:AGENTS.md— no mention of artifact pruning orcargo cleananywhere.REPOSITORY_RULES.md— no mention of artifact pruning.docs/2026-08-09-cargo-artifact-reduction.md— documents profile-based size reduction (debug = 0, strippedciprofile) and a rejected integration-test consolidation experiment, but prescribes no pruning mechanism.docs/superpowers/plans/2026-07-04-fused-elementwise-dag.md:13— contains a plan-scoped note "rundu -s .before work; if over 100GB, propose cleanup before continuing". This is a dated plan instruction for one issue, not a repository-wide documented pruning policy..github/workflows/ci.yml:46-50— CI reportsdu -sh targetanddf -h targetfor visibility but performs no pruning.check-pr*.sh/cleanup script,Makefile/Justfile, or scheduled workflow addresses stale-artifact removal.Expected remediation
Document a concrete pruning mechanism in repository-level documentation (
AGENTS.md,REPOSITORY_RULES.md, or a dedicateddocs/file). Minimum viable options:cargo cleancadence (e.g., weekly, or after every N dependency bumps), and/orfind target -type f -mtime +14 -deleteafter confirming no build is in progress) with usage notes, andtarget/growth is clearly dominated by stale artifacts no current build uses, optionally integrated as a scheduled CI workflow that prunes the runner cache.Scope and exclusions
target/size and ran nocargo build,cargo test,cargo clean, orduontarget/; localtarget/size is owned by a separate parent-AGENTS check and is explicitly excluded here.gh pr list --state openreturns none). No existing issue covers it (gh issue list --state allsearched for artifact/target/pruning/coverage-attestation topics).Policy authority
tensor4all/tensor4all-agent-rules#9— PR that added the build-artifact controls.rules/rust/performance.md:62-65andrules/common/docs-and-tests.md:74-78— the stale-artifact pruning documentation requirements.