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/rome-os/rome/loop-reconcilenpx skills add rome-os/rome --skill loop-reconcilegit clone --depth 1 https://github.com/rome-os/romeWrote 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/rome-os/rome/loop-reconcile)<a href="https://agentmods.dev/skills/rome-os/rome/loop-reconcile"><img src="https://agentmods.dev/badge/skills/rome-os/rome/loop-reconcile.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 5.1 | $0.00037 | $0.01364 |
| Opus 5 | $0.00018 | $0.00682 |
| Sonnet 5 | $0.00007 | $0.00273 |
| Haiku 4.5 | $0.00004 | $0.00136 |
Grade A, and why
loop-reconcile 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 yesterday.
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 — 71 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Loop: reconcile
Goal end-state: every deliverable chunk the run can scope exists as issues, awaiting ready-for-agent, and one report explains whatever it could not scope.
This is the autonomous half of the reconciliation loop. When launched from a session, run it in a background agent. It runs headless: never ask the user a question mid-run. Every ambiguity resolves to a report, never to something plausible. Every report reads standalone on GitHub and never depends on chat context.
All work state lives in the GitHub repository itself: issues, labels, and milestones. No project board. Every issue this skill files carries two labels: loop-reconcile, the mark that the loop owns it, and a scope label loop:<doc-stem> naming its source doc — loop:people for docs/northstars/people.md. A report is an issue too, and it carries loop-complete or loop-needs-discussion on top of those two. If a label is missing from the repository, create it.
Args: an ideal-state doc under docs/northstars/, and optionally a repository milestone.
- Given a milestone, reconcile against the milestone description. First check that the doc still backs every milestone statement: if one contradicts the doc, or the doc no longer asks for it at all, file a
loop-needs-discussionreport and stop. The doc wins on intent, and a milestone that fell behind it needs a session, not work. - Given no milestone, reconcile against the doc's Statements section and file issues attached to no milestone.
- Only the Statements list has authority. The rest of the doc is illustration.
Inputs and precedence
| Source | Role | Authority |
|---|---|---|
| Ideal-state doc | What should be | Wins on intent |
| Codebase | What is | Wins on facts |
| Issue tracker | What is in flight | None — it gates and remembers |
- The tracker never influences what to scope. It only gates whether to scope.
- Do not read PR or git history to guess intent. If a gap looks like a deliberate move away from the doc, scope it anyway. The human withholds
ready-for-agentand fixes the doc. - Concurrency belongs to the driver. Assume this run is alone.
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.
- yesterday Changed · +7 lines · -3 tokens per session 1dced13da78f
- 5d ago First seen · 64 lines · 40 tokens per session scan A b0f29fbfc1c5
loop-reconcile is a skill published in the GitHub repository rome-os/rome (471 stars, last pushed today), licensed MIT. It adds 37 tokens to every session and 1,364 once invoked, about $0.0002 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 2026-08-30.
Other skills, from other repositories
aos-router
Route one inbound user message to exactly one enabled AOS capability. This skill is a strategy contract only. Transport normalization, queueing, permissions, idempotency, audit events, retries, and capability execution stay in Rust.
code-studio-spec
Kiro-style Spec Mode for Code Studio: requirements, design, tasks, implementation, verification, and final report. Runtime sessions, candidate workspaces, diff approval, tests, and cancellation stay in Rust.
watchdog
Parse natural-language WatchDog questions into structured AgentOps queries. This skill documents the intent contract. Rust rules, permission checks, action guards, and fallback answers remain authoritative.
code-studio-code
Cursor/Codex-style coding workspace contract for ordinary Code Mode. This skill documents the behavior expected from the coding agent. Runtime sessions, workspace isolation, process execution, diff validation, and apply/reject actions stay in Rust.
nl2sql-reference
Strategy contract for user-bound reference files in Data Exploration. References can be README files, SQL demos, Python scripts, notebooks exported as text, metric dictionaries, or any UTF-8 text that explains business query patterns. Storage, upload validation, SQL safety, datasource permissions, and query execution…
pm-assistant
Product operations assistant strategy contract. PM short answers and deep analysis should be evidence-grounded and should separate facts from hypotheses. Model calls, hooks, queueing, cancellation, notifications, and staged task execution stay in Rust.