Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Borda/vision-delivery --skill track-and-countgit clone --depth 1 https://github.com/Borda/vision-deliveryWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/borda/vision-delivery/track-and-count)<a href="https://agentmods.dev/skills/borda/vision-delivery/track-and-count"><img src="https://agentmods.dev/badge/skills/borda/vision-delivery/track-and-count/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/borda/vision-delivery/track-and-count"><img src="https://agentmods.dev/badge/skills/borda/vision-delivery/track-and-count.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00151 | $0.01170 |
| Opus 5 | $0.00076 | $0.00585 |
| Sonnet 5 | $0.00030 | $0.00234 |
| Haiku 4.5 | $0.00015 | $0.00117 |
Grade A, and why
track-and-count scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 96 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Produce identity-linked video events—paths, dwell, crossings, or counts—that pass an independently annotated clip-level gate. A detector that works on isolated frames is necessary evidence, not proof that tracking works.
Platform execution boundary. Read ../../resources/roboflow-platform-lookup.md before any provider-specific search, dataset, training, inference, workflow, device, or deployment action. Use installed official Roboflow skills or current MCP resources only for read-only discovery and sourced action-brief inputs; never invoke provider execution. Sentinel owns event semantics and measured delivery evidence.
Follow ../../resources/fde-methodology.md; apply these video-specific rules.
1. Define the event, source, and action
Inspect representative clips, stream configuration, zones/lines, annotations, and code. Ask only what is missing:
- Is the decision about occupancy, unique identity, crossing direction, dwell, or a path?
- What event/action follows, and which miss/false event is worse?
- Is the source recorded video or live stream, and may frames leave the site?
Route instantaneous “how many are visible?” to detect-and-analyze. For live streams, explain the stable delivery choices—local processing, hosted client, or provider-managed runtime—without claiming current availability or commands; exact setup comes from upstream and final integration from deliver-cv-project.
Freeze before tuning:
Acceptance ID: <session/revision>
Business decision: <action enabled by the video event>
Gold set: <independent clip/event labels, sites/times, adjudicator>
Primary metric and threshold: <event recall/precision/count error/dwell MAE>
Secondary guardrails: <identity switches, false events/hour, latency>
Frozen before baseline: <timestamp and confirmation>
Baseline result (diagnostic only): <not run yet>
2. Freeze event semantics
Record coordinate conventions, line direction, polygon boundary behavior, track start/end, minimum dwell, cooldown, re-entry policy, dropped-frame behavior, and source timestamps. Define whether returning objects count again. These rules must not change after inspecting failures without a new acceptance revision.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 96 lines · 151 tokens per session scan A 9a542d61bc65
track-and-count is a skill published in the GitHub repository Borda/vision-delivery (4 stars, last pushed 29d ago), licensed Apache-2.0. It adds 151 tokens to every session and 1,170 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
pipeline
Use when the user wants a feature idea taken end-to-end in one autonomous run — phrases like "run the whole pipeline", "take this feature from idea to finished branch", "brainstorm then build it autonomously", "do everything from idea to merged", "implement all phases without stopping". Triggers when they want…
craft
Use when a product idea is still vague and needs to become a clear definition of what to build — "let's craft an app like X", "help me define what I actually want", "clarify this idea before we plan it". Also use before planning or implementation when requirements, UX, domain behaviour, or technical preferences have…
bug-fix
Use when a bug, regression, or unexpected behaviour is reported and the user wants it fixed end to end — "why is X broken", "this stopped working after Y", "fix this crash". Also use when a symptom is known but its cause is not. Not for building new behaviour, and not for a change whose cause is already proven.
bug-investigate
Use when someone wants to know WHY something is broken and has not asked for it to be fixed — "why is this happening", "what's causing this error", "find out what's wrong", "diagnose this before we decide". Also use before committing to a fix, when the cause is unknown and the decision depends on it. Not for fixing …
setup
Use when superb's skills need their dependencies installed or checked — "set up superb", "install the dependencies", "why does pipeline say superpowers is missing", after a fresh plugin install, or on a new machine. Also use when a skill fails complaining that a superpowers sub-skill cannot be found.
boundedrelay-adaptive-sdd
Coordinate the optional BoundedRelay Spec Kit workflow with strict independent reviews, deterministic effort routing, bounded proposals, convergence, and handoff.