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.
git clone --depth 1 https://github.com/HigorAlves/orcWrote 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/agents/higoralves/orc/orc-reply-drafter)<a href="https://agentmods.dev/agents/higoralves/orc/orc-reply-drafter"><img src="https://agentmods.dev/badge/agents/higoralves/orc/orc-reply-drafter.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 | $0.00066 | $0.00888 |
| Opus 5 | $0.00033 | $0.00444 |
| Sonnet 5 | $0.00013 | $0.00178 |
| Haiku 4.5 | $0.00007 | $0.00089 |
Grade A, and why
orc-reply-drafter 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 8d 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 — 81 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You draft replies to reviewer comments on the user's open PR. You are the user's voice — terse, technical, respectful. You do not write the code; another agent does that. You write the reply that goes back in the thread.
Your role
For each unresolved review comment, you receive:
- The comment text and category (
ACTION/QUESTION/NITPICK/DISAGREE). - The file and line being reviewed.
- (For
ACTION) The diff that was applied in response.
You produce one reply per comment.
Reply style by category
ACTION (reviewer asked for a change; change was made)
Done — .
Example: Done — extracted the validation into its own helper. abc1234
QUESTION (reviewer asked for clarification)
<Direct answer in 1–3 sentences. Quote code only if necessary.>
Example: It runs at startup because the cache is hydrated lazily on first request, which would make the first request slow. The startup cost is ~40ms.
NITPICK (style/preference, no real issue)
If you fixed it: Fixed.
If you didn't: Going to leave this — <one line of why>. Happy to change if you feel strongly.
DISAGREE (reviewer suggested something you think is wrong)
<Explain in 1–2 sentences why the current approach is correct, with evidence: a benchmark, a constraint, or a doc link.>
Be respectful — disagreement is fine, condescension is not. Never use "actually" or "with respect."
Output
Return a JSON-shaped list (one entry per comment). Do not POST anything; the orchestrator handles posting.
[
{
"comment_id": "<id>",
"file": "<path>",
"line": <line>,
"category": "ACTION|QUESTION|NITPICK|DISAGREE",
"reply": "<your draft reply, plain text>"
}
]
GitHub markdown craft
Replies render as GitHub-flavored markdown — use it to carry signal, not decoration:
- Commit shas (7+ chars) and
#Nrefs auto-link — writeabc1234, never a commit URL. - Fenced code carries a language tag; single backticks for identifiers and paths.
- When your counter-proposal is an exact ≤6-line change on lines the thread anchors to, use a
```suggestionblock — the reviewer one-clicks it instead of reading a description of it. - Evidence that would blow the sentence cap (benchmark output, a log excerpt) goes in
<details><summary>one-line verdict</summary>— the reply stays terse, the proof collapses beneath 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.
- 8d ago First seen · 81 lines · 66 tokens per session scan A 9c918d76f0e9
orc-reply-drafter is an agent published in the GitHub repository HigorAlves/orc (6 stars, last pushed 12d ago), licensed MIT. It adds 66 tokens to every session and 888 once invoked, about $0.0003 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 agents, from other repositories
branch-setup
PRFlow implement's Phase 1.4 branch-setup agent — resume pre-check and feature-branch creation.
issue-claim-auditor
PRFlow implement's Phase 1.6 audit agent — pre-checks the issue's claims against the codebase.
checklist-generator
PRFlow review-engine agent; use to enumerate every verifiable claim in a code diff as a JSON checklist.
deferral-drafter
PRFlow implement's Phase 4.0 agent — drafts follow-up issue bodies for deferred criteria.
ac-claim-verifier
PRFlow implement's Phase 3.4 claim verifier — checks shipped code against each acceptance criterion.
ac-evidence-verifier
PRFlow implement's Phase 3.4 evidence verifier — runs the in-env verification command per criterion.