[blankym]

Problem and outcome

Standalone UnityFS AssetBundles were opaque artifacts. The existing CLI and MCP inspection/extraction workflows now enumerate and recover format 6–8 members without running Unity. Version 9 has a different block-table layout and remains unsupported.

Approach and contract

• One parsed header/table owner preserves producer-specific alignment and encryption flags, raw node flags, nullable unknown facts, and virtual offsets. Virtual member offsets are not physical carrier locations (source_location: null).
• Directory buffers/rows/path projections are admitted before allocation. Member reads have separate plaintext and I/O-work budgets, bounded active streams, directly streamed plain ranges and one immutable completed-block cache. Empty-output rows retain their source count/offsets but do not multiply member traversal work.
• Streams remain owned through decoder/I/O retirement, cancellation and close; failed descriptor cleanup remains retryable. Carrier freshness is checked at member open and completion.
• Producer-owned content availability and traversal coverage remain distinct. Unsupported UnityFS members retain their names/sizes with unavailable hashes and explicit unknown coverage; ASAR missing companions preserve the existing per-entry semantics. Generic shutdown/cleanup failures propagate.
• Byte-based artifact detection, shared reader construction, desktop header probes and existing retained filtered/paged views use the same contracts. Caller-selected extraction outputs use the existing output ownership/cleanup workflow; source bytes are preserved.

Executed validation

• Source, compiled CLI, full/compact stdio MCP, Godot PCK, ASAR cleanup/identity and desktop checks ran locally. Latest focused workflow run: 42/42; strengthened empty-block/I/O/lifecycle suite: 17/17. Existing Next-context acceptance cases pass 4/4 after distinguishing availability from traversal coverage; the earlier change reproduced 2 failures while the unchanged matching-main baseline passed 4/4.
• npm run check:fast, npm run docs:check, changed-file formatting and bounded single-thread typed lint passed. Final-head CI run 40656542039 passed, including all eight portable shards and CI required; Greptile review is 5/5 with no unresolved threads. Merged as e78ac1a.
• Public UnityPy corpus pinned at 417d59522a998fdee9a9c2b29fe35f3063222457: atlas_test, banner_1 and char_118_yuki.ab have six raw members verified by independent UnityPy 1.24.2 SHA-256 and actual CLI/MCP extraction. atlas_test source SHA-256: 61ee5afb5f5c64f0a4f65f039d09b2509198363e713423289c895067ac215177.
• Public xinzexi_2_n_tex contains LZMA data: both names/sizes remain visible with unavailable hashes. Five independent UnityPy repacks additionally cover format 7/8, front/end tables, padding and unsupported LZMA data/table behavior. These are library-produced files, not Unity engine executions.
• Real compact MCP retained views return about 911–1017 estimated tokens for selected facts versus 2130–2301 for complete inspection responses. The entire four-file verification session, including catalog and text/structured response duplication, was about 87,397 estimated tokens (catalog 50,939); tokenx estimates are not billing counts.
• Final 94fab716c was compared with the original reader on the same main (2b8fb387b, baseline merge fa5296412): 12 alternating completed inspect/extract/SHA/cleanup tasks over atlas_test and char_118_yuki.ab, three fresh-process repetitions per file/lane. Before → after: elapsed 4.99–6.06 → 5.11–6.05 s; CPU 6.52–7.11 → 6.73–7.10 s; sampled process-family RSS 260.5–282.0 → 262.2–280.0 MiB. Request+stdout estimates 3253/3305 → 3237/3289 tokens. First/repeat runs do not imply flushed OS caches; ranges overlap, so no general speedup is claimed.
• A completed final four-file MCP task measured 8.80 s elapsed, 10.86 s CPU and 452.4 MiB sampled process-family RSS, including SDK client, server, catalog, selected views, SHA checks, cleanup and measurement/reporting. CPU uses wait4 plus sampler CPU; family RSS is sampled every 10 ms. Node 24.22.1 heaps were capped at 2 GiB, test workers at two. See UnityFS bounds and scope.

Verification scope excludes real Unity engine execution, version 9, decryption, SerializedFile object/texture decoding, and embedded-carrier discovery. Input hashes, extraction cleanup receipts, per-task elapsed/CPU/family-RSS and token estimates are retained locally; no binary fixtures or planning artifacts are committed.

Author readiness

Human owner: @blankym

• I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
• I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
• I declare this PR ready for initial maintainer review.

Before merge

• This PR addresses a concrete problem or an agreed enhancement.
• Material scope and authority questions are resolved, and any required maintainer scope decision covers the final diff.
• The PR has one focused outcome and the title follows type(scope): outcome.
• Related issue is linked, or the reason for not linking one is stated above.
• Tests cover changed observable behavior and meaningful failure paths.
• Owning docs, contracts, and generated metadata are updated where needed.
• User-visible CLI/MCP changes include representative output.
• I checked the final diff for secrets, unrelated cleanup, and unsupported claims.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-28T02:33:35.059388Z We couldn’t complete this review. Comment @codex review to try again.	0dcac0e	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The reviewed changes appear safe to merge, with both numbered findings addressed and no new blocking issue found.

Summary

Adds standalone UnityFS format 6–8 inspection and extraction through the shared CLI/MCP workflows.

• UnityFS bundles expose readable members without running Unity.
• UnityFS bundles enter the existing inspection and extraction workflows.
• Unavailable member bytes stay separate from reader and cleanup failures.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[CLI or MCP input] --> B[Byte-based detection]
  B --> C[UnityFS reader]
  C --> D[Bounded header and directory parsing]
  D --> E[Named members and source facts]
  E --> F{Content supported?}
  F -->|Yes| G[Bounded member stream]
  G --> H[Hash or extract bytes]
  F -->|No| I[Unavailable hash and unknown coverage]

Loading

_{Reviews (3) · Last reviewed commit: "fix(artifacts): preserve coverage author..." · Reviewed by Greptile}