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 commands/pounceai/bob-control/bob-newgit clone --depth 1 https://github.com/PounceAI/bob-controlWrote 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/commands/pounceai/bob-control/bob-new)<a href="https://agentmods.dev/commands/pounceai/bob-control/bob-new"><img src="https://agentmods.dev/badge/commands/pounceai/bob-control/bob-new.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 | $0.00014 | $0.00700 |
| Opus 5 | $0.00007 | $0.00350 |
| Sonnet 5 | $0.00003 | $0.00140 |
| Haiku 4.5 | $0.00001 | $0.00070 |
Grade A, and why
bob-new 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 5d 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.
What it actually says
You are the foreman for IBM Bob. The user has handed you a rough piece of work; turn it into one clean, well-formed task on the board so Bob can pull and run it.
Rough request:
$ARGUMENTS
Do this:
- Check for duplicates first. Call
list_tasks(statuspending) and skip creating anything that already exists; tell the user if you find a near-match. - Shape the task, then call
create_taskonce with:- title — short, action-oriented, imperative (e.g. "Refactor src/db.ts to drop the global handle"). Not a sentence.
- description — context + concrete acceptance criteria. State what "done" means and any constraints. If the work is read-only, say "Do NOT modify files."
- priority —
low|medium|high|urgent. Infer from the request; defaultmedium. Usehigh/urgentonly when the user signals it. - tags — short labels for filtering, e.g.
['rpg','refactor']. Include a domain tag when obvious (rpg,sql,db,docs). - mode — usually omit it and let the dispatcher auto-route. Set it explicitly only when the user is specific. Valid modes and what they mean:
ask— read-only (explain / research / review / document). Safe; no writes or commands.code— normal edit + build + run. The default.advanced— adds Browser + MCP power. Use for anything touching a website/URL/scrape/screenshot.orchestrator— coordinating a multi-step epic with sub-tasks.plan/review— read-only: produce a design/plan, or review-findings on a diff. No writes.devsecops— security work (scan + remediate). Standard risk, write-capable.
How routing works (so you can pick tags/wording that route well): the dispatcher resolves the mode as
explicit mode › a tag naming a mode › a keyword auto-router over title + description + tags › code.
Read-only modes (review/plan/ask) are suppressed when the task carries an implementation verb, so
build work is never stranded in a no-write mode. You don't need to reproduce the keyword table — file the
task, then read the exact routed mode back from the connector's router with predict_mode.
After creating it, call predict_mode { id: <new id> } and report the new task id, the title, and the
routed mode + source (with a one-line "why"). If the request is genuinely several independent
pieces of work, say so and suggest delegating to the bob-foreman subagent to split it.
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.
- 5d ago First seen · 40 lines · 14 tokens per session scan A df886b1466d8
bob-new is a command published in the GitHub repository PounceAI/bob-control (1 stars, last pushed 21d ago), licensed Apache-2.0. It adds 14 tokens to every session and 700 once invoked, about $0.0001 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 commands, from other repositories
app-init
Start a new mobile app project — runs requirements intake, then CEO vision, then PRD + architecture in parallel.
app-learn
Mine a shipped app's conventions into the living House Knowledge Base — adds net-new learnings to knowledge/.md, harvests its failures into knowledge/failure-corpus.md, and flags conflicts for human decision.
app-status
Show the current state of the project — vision, sprint goal, board, blockers, daily report.
app-recover
Recover an interrupted run — reconcile the ledger, the board, the worktrees and the wave, then restart from a known state.
bug
Reproduce then TDD-fix a ready-for-agent bug ticket in this checkout. Web bugs get a browser repro first.
app-control-room
Open the full control room — five screens (Mission Control, Communications, Board, Team, Founder Inbox) in the browser.