v-modal/vmodal_mcp / skills/vmodal-track
vmodal-track
VerifiedTrack a subject through VModal video collections: find every verified appearance of it and return a trail (where, when, for how long, in what order). Works for a description ("red taxis in Kyoto") or for one specific subject ("this taxi", "the same woman", "the one at 1:57"). Use when the user asks to track, follow, trace, or find everywhere / every time something appears, e.g. "Track red cars. Collection: city". For a single good match use vmodal-find instead. Requires the vmodal MCP (dev branch) and the vmodal-find skill.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 8.1 KB |
| reference.md | markdown | 3.0 KB |
| scripts/anchor_b64.shexecutable | script | 1.6 KB |
reference.md
View on GitHub# vmodal-track reference
Shared rules (spec checklist, scenes, verdicts, timestamps, frame tools) are in
[../vmodal-find/reference.md](../vmodal-find/reference.md). This file covers what tracking adds.
## Budgets
| Item | Value |
|---|---|
| text queries | 3: full, core + distinctive attribute, one rephrasing |
| image queries (identity mode) | 1, at most 2 |
| hits per query | MCP caps a call at 20 rows: `limit=20`; page 1 of each query + page 2 of the full query, then more pages (`offset` 40, 60, 80) only when all found appearances are viewed and the stop rule allows; max 5 pages per query |
| scene gap | ≤ 3 s (same as vmodal-find) |
| appearance gap | ≤ 10 s between scenes of the same video |
| second frame | last hit of a FAIL appearance spanning ≥ 5 s; vmodal-find second look for unclear PARTIALs |
| batch | 10 appearances per bulk frame fetch |
| stop | 2 batches in a row with no new PASS / SAME, or 50 appearances viewed |
## Search by image
- Tool: `search_video(group_name, stream_name?, image_query=<base64>, limit=20)`. `find` has
no image input.
- `image_query` must be **plain base64 image bytes**. A signed URL fails with
`400 cannot decode image_query base64: Incorrect padding`.
- Make it with `bash <skill_dir>/scripts/anchor_b64.sh <frame.png> x y w h`. Frames are
640×360; take the box from the frame you viewed, with a small margin around the subject.
- **Crop the subject.** Tested on `city`: the whole Kyoto taxi frame (1:57) returned mostly
the same 1:54–2:05 stretch of street. The cropped taxi also returned other Kyoto scenes
(2:25, 2:29, 4:24, 6:14, 7:45) and two Rome videos.
- Scores from image search are a rank too: use the returned order, not the number.
## Identity labels
| Label | Rule | On the trail? |
|---|---|---|
| SAME | visible signature details all match; ≥ 1 clearly visible | yes |
| LIKELY SAME | no contradiction, key details hidden | yes, marked "likely" |
| DIFFERENT | a signature detail contradicts | footer count |
| DESCRIPTION ONLY | fits the description, nothing tells it apart | footer count |
Good signature details: plate or number, logo, text on the subject, pattern, accessory,
damage, unusual color combination. Weak ones: color alone, type alone, location.
## Worked example (identity mode)
Request: `Track the taxi with plate 50-20. Collection: city, Kyoto`
1. Anchor: `find("red taxi with a white roof", "city", "Street_Visit_in_Kyoto")` → 1:57.
Frame shows plate 50-20, white roof light, sticker on the rear window.
2. Signature: `plate 50-20` · `white roof light` · `rear window sticker`.
3. Search: 3 `find` variants + one `search_video` with the taxi cropped at x=225 y=190 w=210 h=160.
4. Group → appearances, e.g. 1:52–2:05 (one appearance, several scenes), 3:31, 4:24.
5. Verify: 1:52–2:05 SAME (plate readable); 3:31 taxi drives past, plate not readable,
other details don't match → DESCRIPTION ONLY or DIFFERENT, never SAME.
6. Stop after two batches with no new SAME; report the trail and the footer counts.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |