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 agentmods add skills/broomva/skills/handbacknpx skills add broomva/skills --skill handbackgit clone --depth 1 https://github.com/broomva/skillsWrote 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/broomva/skills/handback)<a href="https://agentmods.dev/skills/broomva/skills/handback"><img src="https://agentmods.dev/badge/skills/broomva/skills/handback.svg" alt="Measured on agentmods" 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 | $0.00275 | $0.03670 |
| Opus 5 | $0.00138 | $0.01835 |
| Sonnet 5 | $0.00055 | $0.00734 |
| Haiku 4.5 | $0.00028 | $0.00367 |
Grade A, and why
handback 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 4d 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 — 230 lines — stays where its author put it; the contents beside it link to each section on GitHub.
handback — end an arc with a work order, not a lab notebook
Spec: broomva/workspace → docs/specs/2026-08-18-agent-handback-contract.html (BRO-2179).
Why this skill exists
Measured over 994 Claude Code transcripts (analysis set: 100 sessions of ≥6h wall-clock and ≥150 assistant turns):
| Long arcs whose final message mentions a blocker at all | 31 / 100 |
| Long arcs halting on a human (terminal-stance phrasing) | 18 / 100 |
| …of those 18, containing a conforming ask block | 0 / 18 |
| …containing an imperative anywhere in the message | 3 / 18 |
…containing any ? at all |
13% |
| …with the question next to the blocker it describes | 6% |
| Stating a default if the human stays silent | 5% of all 100 |
| Ever emitting a push notification | 0 / 100 |
| Blockers knowable before the arc started | 81% |
The two blocker rows differ on purpose. The broad count (31) includes messages that merely mention being blocked — "the pre-commit hook blocked it", "#381 remains blocked by the hang". The narrow count (18) is the defensible one: a terminal message that stops the arc on a person. The enforcement gate keys on the narrow definition, because refusing a healthy receipt that happens to contain the word "blocked" would fight the operator.
The ask usually exists. It is placed last, written in the indicative ("Not merged — auto-merge correctly blocked"), and carries no default, so the arc dies on it — median 2.4h of stalled arc-time behind such a message, 25 of them overnight-length.
Rule zero — asking is the last resort, not the interface
No ask may exist until every rung below has been tried and recorded. A row
with an empty exhausted list is a defect, not a question.
| Rung | Try this | |
|---|---|---|
| 1 · Read | Is the answer already on disk — ticket history, prior conversations, the knowledge graph, the handoff doc, the code? | Most "decisions" are recoverable, not new |
| 2 · Standing grant | Has this exact question been answered before? Check .control/preauth.yaml |
Re-asking a settled question is the cheapest failure to eliminate |
| 3 · Delegate | Can a fresh agent resolve it — a reader, a researcher, an adversarial reviewer with different context? | Chain agents instead of chaining to the human |
| 4 · Reroute | Can another lane, worktree, fan-out, or approach go around it? | The blocker is often a path, not a wall |
| 5 · Loop | Can iteration resolve it — retry under a changed assumption, let a persist loop converge? | Slow beats blocked when nobody is awake |
| 6 · Default | Is the choice reversible? Take it, log it, tell the human afterwards. Do not ask. — "log it" means an entry in the ledger's decisions:, and "tell the human" means the 🔀 block below |
Reversible decisions are not the human's to make in real time |
| 7 · Ask | Only now. Record which rungs were tried and why each failed. |
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 4d ago First seen · 230 lines · 275 tokens per session scan A a2c96ae7ca67
handback is a skill published in the GitHub repository broomva/skills (3 stars, last pushed 4d ago), licensed MIT. It adds 275 tokens to every session and 3,670 once invoked, about $0.0014 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
cococonverge
CocoPivot — multi-pod convergence synthesizer. Deterministically deduplicates findings from N parallel pod outputs into one prioritized, coordinated action plan. Invoked via $pivot run/show/status/clear.
crewai-multi-agent
Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when you need role-based agent collaboration with memory, or for production workflows requiring sequential/hierarchical execution. Built without LangChain dependencies…
ray-train
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.
skypilot-multi-cloud-orchestration
Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.
background
Use when the user wants to see, inspect, cancel, or prune background agents fired during prior chain runs. Read/manage .hyperflow/background/registry.json and the per-agent output buffers at .hyperflow/background/ .md. Standalone — never auto-invoked. Trigger with /hyperflow:background, "list background agents"…
local-frontend-check
Smoke-test or verify UI behaviour on the local Jarvis Registry frontend running at http://localhost/gateway. Use for manual regression checks, bug-fix verification, and end-to-end confirmation of specific flows without running the automated test suite.