Session Prompt — WP-003 — 2026-06-02

Paste this as the first message to the executing Claude Code session. The session opens execution for WP-003 against the cached TGIFILE.ART.

Header

• WP: WP-003
• Title: Inspect the 1 MB pre-payload region of TGIFILE.ART (palette hunt)
• Session date: 2026-06-02
• Paired pre-flight: pre-flight-WP-003-2026-06-02 — verdict READY, signed against commit 0eaad13
• Targets generation(s): Gen 1 (Showdown only)
• Targets disc(s): Showdown — operating against the cached TGIFILE.ART, no ISO re-mount needed

Intent

Characterize the byte range 0x0F48–0x10017E (1,045,046 bytes) of TGIFILE.ART: palette table, asset name table, frame metadata, padding, or some combination. Resolve the 4-byte gap at 0x0118–0x011B (between the declared group descriptor end and the asset count field, currently labeled in tgifile-art.md). Land structured Findings under tgifile-art.md that downgrade the “packed/padded region” label to concrete content — or, if the region really is padding, characterize why (alignment boundary, sector-multiple, etc.) with byte-citation evidence.

This unblocks WP-001 (Ghidra session can drop palette-hunt scope from its TGIFILE.ART decode trace) and gives WP-002 a head-start on first-decode strategy (palette-indexed if a palette is here).

Explicitly NOT in scope for this session: any Ghidra work, any payload decoding (≥ 0x10017E), any changes to format docs other than tgifile-art.md, any Phantom claims (Phantom’s archive format is unverified per scooby-exe §Pending: Phantom archive format — Phantom-specific findings are deferred to WP-006).

Authority chain (read order)

Read these files top-to-bottom before doing any work:

1. C:\Users\jjensen\.claude\CLAUDE.md — operator preferences (tone, ownership, even-handedness, commercial defaults)
2. docs/01-VISION — engine-lineage finding, generation discipline
3. docs/reference/wp-lifecycle — Phase 3 execution discipline + commit hygiene + the §End-to-end operator workflow steps 8–13
4. Paired pre-flight artifact: pre-flight-WP-003-2026-06-02 — Source binary identity table + scope lock + risk review
5. WP body: WP-003-pre-payload-region — §Scope + §Exit criteria + §Notes (hand-off from WP-007 + generation discipline)
6. Format doc to update: docs/formats/tgifile-art — current state of the spec; the section labeled “packed/padded region” is the one this WP downgrades or replaces
7. Cross-check inputs: docs/formats/scooby-exe for Showdown’s string-keyed asset resolution pattern; tools/exes/showdown/strings-ansi.txt (gitignored, locally accessible) for candidate-name cross-check
8. Cache reference: tools/exes/README §Showdown — TGIFILE.ART cached identity (SHA-256 lock)

Pre-execution checks

Confirm each before touching files:

• git status --short returns empty
• HEAD on main at or after commit 0eaad13 (“docs: WP-003 step-4 revision — lifecycle fields + WP-007 hand-off + cache pin”)
• Paired pre-flight artifact pre-flight-WP-003-2026-06-02.md is on main with verdict READY
• TGIFILE.ART SHA-256 matches the Source binary identity table in the pre-flight:
  • Get-FileHash tools\exes\showdown\TGIFILE.ART -Algorithm SHA256 → B3006E127B4EFA19CF419E92A97A8AA0A565378243CBE396ABA9707F3310C807
• Python 3 available (py -3 --version — use the py -3 launcher, not bare python, per the WP-007 pre-flight lessons-learned note about Inkscape’s bundled Python on PATH)

If any check fails: STOP. Surface the failure; do not run hex inspection against mismatched state.

Execution rules

Scope allowlist (verbatim from pre-flight §Scope lock):

docs/formats/tgifile-art.md
docs/reference/pre-flight-WP-003-2026-06-02.md
docs/sessions/session-WP-003-2026-06-02.md
docs/work-packets/WORK_INDEX.md
docs/work-packets/WP-003-pre-payload-region.md

Allowed but local-only (gitignored under tools/samples/*):

tools/samples/tgifile-art-palette-0.bin
tools/samples/tgifile-art-prepayload-dump.bin

Forbidden:

• Game data binaries committed to git — .gitignore is the safety net; do not bypass
• engines/scooby/** — out of scope
• docs/formats/ files other than tgifile-art.md
• Any docs/execution-checklists/ edits — none are paired with WP-003
• Reading byte ranges outside 0x0F48 through 0x10017E of TGIFILE.ART. Payload bytes (≥ 0x10017E) are WP-002 scope; header / asset entry table (0x0000–0x0F48) bytes may be read only for sanity-check cross-references to facts already locked in tgifile-art.md.
• Anything outside the allowlist above

Generation discipline (REQUIRED): every Finding subsection labels its source as #### Gen 1 (Showdown). Phantom claims are forbidden — Phantom’s archive format is unverified (WP-006 territory). Cross-generation claims explicitly span (or explicitly stay scoped to Gen 1).

Copyright posture: local-only palette bin and prepayload dump under tools/samples/ are gitignored; never staged. Into docs/formats/tgifile-art.md go ONLY: byte offsets, sizes, hex-window citations, extraction recipes, and characterization prose. Do not paste raw hex blocks longer than ~64 bytes; cite offset + length + a short representative window.

Commit conventions:

• docs: prefix; imperative subject under ~80 chars
• Two-commit topology (per wp-lifecycle §Two-commit topology, even though this is docs-only — the close is a separable review surface):
  • Commit A: docs: WP-003 — characterize TGIFILE.ART pre-payload region
  • Commit B: docs: close WP-003 — flip to ✅ Done
• Body explains the why in 1–4 sentences
• Co-Author footer:

  Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
  

Top risks (imperative form from pre-flight §Risk review):

• You MUST cite the SHA-256 B3006E127B4EFA19CF419E92A97A8AA0A565378243CBE396ABA9707F3310C807 of TGIFILE.ART in the Findings header. A finding without the SHA-256 anchor is invalid.
• You MUST stop hex reads at offset 0x10017E exclusive. Any byte cited at or above 0x10017E is a scope violation; that’s WP-002 territory.
• You MUST characterize the FULL byte range 0x0F48–0x10017E, not just the first thing found. Even if a palette is identified at 0x0F48 and ends at some offset X, the bytes from X through 0x10017E still need characterization (continued palette? padding? second table?).
• If candidate asset-name strings turn up, you MUST cross-check each against tools/exes/showdown/strings-ansi.txt (gitignored, local) per the WP-003 §Notes hand-off. Record the cross-check result for each candidate.
• If a palette is extracted to tools/samples/tgifile-art-palette-0.bin, the Findings entry MUST include the byte-offset extraction recipe (start, length, post-processing if any) so a future operator can regenerate the bin from the SHA-256-locked TGIFILE.ART. The bin is local-only; the recipe is the committed reproducibility surface.

Session task

Execute WP-003 §Scope in this order:

1. Sanity-check the cached binary. Read the first 16 bytes of tools/exes/showdown/TGIFILE.ART and confirm the group count (uint32LE at offset 0x0000 = 69) and the asset entry count (uint32LE at offset 0x011C = 453) match the values currently locked in tgifile-art.md. If either differs, STOP — the cached file is not what the format doc expects; the SHA-256 lock alone may be insufficient.

2. Resolve the 4-byte gap at 0x0118–0x011B. Read these 4 bytes in hex. Likely outcomes: (a) zero padding (record as such), (b) unidentified field (record the hex value verbatim; future Ghidra session may explain it), (c) the asset entry count is actually at 0x0118 not 0x011C and tgifile-art.md has the offset wrong by 4 (record + update the spec). Document in tgifile-art.md Findings under a small “4-byte gap at 0x0118” subsection.

3. First pass — region characterization (high-level). Read the full 1,045,046-byte range 0x0F48–0x10017E into memory (Python Path.read_bytes + slice, or any hex viewer’s go-to-offset + select). Compute:
   • Byte-value histogram — long runs of 0x00 indicate padding; uniform distribution suggests compressed or structured data
   • First non-zero byte after 0x0F48 (the structure-start offset)
   • Last non-zero byte before 0x10017E (the structure-end offset)
   • Any periodic structure — try strides of 3, 4, 6, 768, 1024 (palette candidates) and 8, 12, 16 (entry-table candidates)
4. Targeted hypothesis checks in order, each citing offset + hex window + verdict:
   • Palette hypothesis A — 256-entry RGB (768 bytes): is there a contiguous 768-byte run at any offset in the region where bytes look like color values (no extreme outliers, plausible gradients)?
   • Palette hypothesis B — 256-entry RGBA (1024 bytes): same check at stride 4
   • Palette hypothesis C — multiple palettes: does the region contain multiple 768-byte or 1024-byte blocks at regular strides?
   • Name table hypothesis: does the region contain ASCII strings ≥ 6 chars? If yes, sample a handful and cross-check against tools/exes/showdown/strings-ansi.txt (WP-007 dump). A candidate appearing in both is a strong signal; appearing only in TGIFILE.ART is a weaker signal (the name table might be data-local rather than code-referenced).
   • Frame metadata hypothesis: is there entry-shaped data (fixed-stride records) that could be per-asset metadata (width, height, palette index, frame count)?
5. Author Findings in tgifile-art.md per WP-003 §Exit criteria:
   • Replace or downgrade the “packed/padded region” label with concrete characterization
   • One subsection per identified structure (palette, name table, metadata, padding), each citing byte offset + length + hex window + cross-check result
   • The full byte range 0x0F48–0x10017E must be accounted for — no gaps in the characterization. If a sub-region is genuinely padding, say so and characterize why (alignment boundary, sector-multiple, etc.) with byte-citation evidence.
   • If a palette is identified: extract to tools/samples/tgifile-art-palette-0.bin (local-only); record the byte-offset extraction recipe + the bin’s SHA-256 in the Findings entry.
   • The 4-byte gap at 0x0118 is resolved (step 2 above).

6. Commit work as Commit A (docs: WP-003 — characterize TGIFILE.ART pre-payload region).

7. Update WORK_INDEX — flip WP-003’s Status to ✅ Done. Annotate downstream rows (WP-001 EC-001 palette-hunt scope reduction; WP-002 first-decode strategy if palette was found). Update §Current state pointer. Commit as Commit B (docs: close WP-003 — flip to ✅ Done).

Outputs this session MUST produce:

1. Updated docs/formats/tgifile-art.md with Findings characterizing the full 0x0F48–0x10017E range, the 4-byte gap at 0x0118 resolved, SHA-256 of source TGIFILE.ART cited
2. Updated docs/work-packets/WORK_INDEX.md with WP-003 row at ✅ Done + downstream annotations
3. Updated docs/work-packets/WP-003-pre-payload-region.md body with Status → ✅ Done, actual effort recorded, Findings-landed links
4. Updated docs/reference/pre-flight-WP-003-2026-06-02.md §Lessons learned (one line or “No lessons learned”)
5. Local-only (gitignored): tools/samples/tgifile-art-palette-0.bin if a palette is found

Outputs this session MUST NOT produce (sprawl guard — log as follow-up candidates instead):

• Ghidra annotations or .gpr work (WP-001)
• Payload decoding (WP-002) — any read at or above 0x10017E is a scope violation
• Phantom-specific claims (WP-006 territory)
• Format-doc updates outside tgifile-art.md — even if the strings cross-check surfaces a finding that touches scooby-exe.md, leave that for a follow-up
• Edits to EC-001 — the WP-003 / WP-001 hand-off is documented in WP-003 §Notes; EC-001 itself isn’t touched here

Close ritual

When all WP-003 §Exit criteria are satisfied:

1. Confirm outputs landed. Every file in the §Scope lock allowlist has the expected change. Verify the Findings section in tgifile-art.md characterizes the full byte range with no gaps, the 4-byte gap at 0x0118 is resolved, and (if applicable) the palette extraction recipe is documented in enough detail to regenerate the local bin.
2. Confirm raw extracts gitignored. git diff --cached --name-only shows zero tools/samples/** content files.
3. Verify the two-commit topology is on main (or on a branch if one was used — direct-to-main is the WP-007 precedent for docs-only WPs). Commit A = findings; Commit B = WORK_INDEX flip.
4. Propagate to dependent WPs. WP-001 and WP-002 both cite WP-003 as a “recommended” prerequisite — note in WORK_INDEX §Current state that the WP-003 baseline is now landed, and if a palette was found, update WP-002’s first-decode-strategy line directly.
5. Update the lessons-learned section of the pre-flight artifact — either “No lessons learned” if execution was clean, or one line on any preventable gap pre-flight should have caught.
6. Push to origin/main (per the wp-lifecycle triage off-ramp fix — direct-to-main commits don’t squash-merge themselves; git push origin main is required for GitHub Pages to rebuild).
7. Report back to operator. State: outputs landed, exit criteria satisfied, status flipped, pushed. Surface any anomalies that came up but didn’t fit existing WP/EC scope as follow-up WP candidates.

If any §Exit criterion is NOT met: WP-003 stays 🚧 In Progress. Do NOT flip WORK_INDEX. Document what’s open in WP §Notes and report back honestly.