security: bound archive member enumeration before materialization

[seonghobae]

Replacement stack repair

This draft supersedes now-closed conflicted/stale-stack PR #70. It reconstructs the same parser-resource hardening on the exact current replacement predecessor #86 without rebasing, retargeting, force-updating, or rewriting either historical branch.

• Exact predecessor security: bound direct SBOM archive input before parser execution #86 head/base of this PR: df267fe522c56a28b6ec7d9679bf5edd710b4200.
• Immutable test-only RED head: 22d3d66ac3f5ff35b100a4095aa09afb07fa0da1.
• Exact current implementation head: 0f6476ee4f705da7b0cd5fa7120b87dcea5803b6.
• Closed security: bound archive member enumeration before materialization #70 remains immutable TDD/review history only; no check or review from security: bound archive member enumeration before materialization #70 is reused as acceptance here.

Immutable RED evidence

Exact-head CI run 31150692931 checked out 22d3d66ac3f5ff35b100a4095aa09afb07fa0da1 with persisted credentials disabled. Package acceptance and Ruff succeeded. Every Python 3.10–3.13 test lane failed at the focused regressions; Python 3.13 reported 15 failed, 775 passed.

The RED tree proved the exact predecessor still entered ZipFile/tar parser materialization before enforcing the 10,000-member ceiling and lacked bounded ZIP central-directory, expanded-tar, and extension-header preflight controls.

Reconstructed bounded implementation

The exact current tree reapplies the reviewed #70 resource-boundary delta onto #86's later descriptor/path-identity hardening rather than copying the stale branch wholesale.

• scans one canonical single-disk non-ZIP64 ZIP central directory with bounded descriptor reads before ZipFile allocates ZipInfo objects, requiring EOCD/record count/size/offset agreement and enforcing at most 10,000 members;
• streams gzip/tar physical headers before semantic parsing, enforcing at most 10,000 physical members, a 512 MiB aggregate expanded-tar ceiling, and a 1 MiB per-extension-header ceiling;
• rejects malformed/truncated central directories, multi-disk/ZIP64 forms, malformed ZIP extras, invalid tar framing/checksums, links, devices, FIFOs, sparse archives, unsupported forms, nonzero trailing data, and decompression corruption through stable fail-closed errors;
• uses sequential tar parsing rather than getmembers() materialization;
• preserves security: bound direct SBOM archive input before parser execution #86's no-follow preflight, descriptor/path identity, finite compressed-byte ceiling, live read/seek/position bounds, growth/mutation rejection, metadata/dependency semantics, deterministic CycloneDX output, and stable direct-artifact errors;
• keeps scripts/ci inside the canonical production coverage scope and retains only directly relevant operator/research documentation and [Unreleased] evidence;
• changes no .github workflow, credential, permission, model, network transport, protected ref, publication, signing, attestation, or release authority.

Exact-current-head verification

CI run 31151452927 checked out exact SHA 0f6476ee4f705da7b0cd5fa7120b87dcea5803b6 explicitly and succeeded across Python 3.10, 3.11, 3.12, and 3.13 plus wheel/sdist package acceptance. Python 3.13 recorded:

• 790 passed;
• production coverage exactly 1,631/1,631 statements and 558/558 branches (100%, zero partial branches);
• release_evidence.py at 371/371 statements and 122/122 branches;
• Ruff, compileall, and hourly product-guard self-test success.

Because this PR remains draft behind #86 → #64 → #62, draft-skipped SAST/Security/review evidence is not acceptance. The organization Security Scan must also be rerun after ContextualWisdomLab/.github#799 integrates; its reviewer prerequisite is .github#813.

Remaining gates

Keep this PR draft while predecessor #86 is draft and transitively depends on #64/#62. Before integration require accepted predecessors, exact-current-head applicable SAST and repaired organization Security Scan, completed automated review, qualifying independent non-author approval, zero unresolved valid threads, branch protection, and every repository policy gate. Missing, pending, draft-skipped, stale-head, predecessor-head, synthetic-merge, rate-limited, or fail-open evidence is not acceptance.

Do not merge, auto-merge, release, publish, rebase, retarget, or force-update while draft.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 50577628-1086-415b-a87c-7ea5ffd26c2f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}