메뉴
Pace PR creation — one PR open at a time per repo, one per org at a time, address feedback before pushing new. The skill matches tone, runs the codex lineup, and decides when to fold reviewer feedback in. Queue state, hard gates, gh I/O, and the heartbeat live in `sweep` CLI.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Filter recent Hacker News for live threads on June's topics and hand him the list. Separately, flag the rare subset where work he has already done settles a specific quoted claim, and draft that comment as a plain argument with no link to his own writing. Use when June asks to check HN or find threads worth reading or commenting on.
Scan blog posts for AI writing tics and missed opportunities for human voice. Reports subtractions (AI patterns to remove) and additions (wordplay, arc, and claims to strengthen).
Poll June's Hacker News feeds, report what is new since the last run, and idempotently arm a background watch for replies to his comments. State persists in ~/Documents/sweep/ so it survives across sessions. Use when June asks what is new on HN, whether anyone replied, or to start watching a thread.
Argument-level compression by a strengthened absence test (does the argument stand with the same force, clarity, evidence, scope, pacing, and misreading-resistance?). Removes passages that fail it (self-referential cruft, cross-section restatement, self-recap, re-derivation, scaffolding, expired reader-orientation, stakes inflation, inert hedges, redundant examples); grafts good rhetoric onto an adjacent essential point rather than deleting. Idempotent. The structural cousin of /tighten (word-level) and /not-but (negation).
Audit a technical candidate's resume/CV by verifying its claims against the public contribution graph instead of taking the text at face value. Sorts every claim onto the self/peer/world attestation lattice, applies a cost-and-stake filter so cheap proxies (stars, downloads, backlinks) don't count, runs LIVE checks (GitHub, arXiv, package registries, CVEs), and reports two separate axes — merit (staked acceptance of the candidate's specific work) and attention (timestamped output) — plus a verifiability score in [0,1]. For technical recruiters and hiring managers. Invoke when given a candidate — a GitHub handle, a resume/CV, an arXiv author, a personal site — and asked to verify, audit, vet, or check their claims. Runs at two depths — a quick triage scan or a deep per-claim audit — and elicits which. Checks the provenance of a personal site or self-hosted resume (Internet Archive + git history) as an un-backdatable anti-tailoring anchor. Needs tool access (Bash/curl or WebFetch).
Find and defuse "not X but Y" / "isn't X; it's Y" / "X — not Y" constructions. Deterministic grep for detection (no LLM blind spot), the not-but subagent for triage. Split-or-cut tail negations ("X, not Y") aggressively — promote the point to its own sentence or delete it; recast in place only mid-clause "not X but Y" that genuinely can't be split; keep just the rare iconic thesis line literal.
| name | drip |
| description | Pace PR creation — one PR open at a time per repo, one per org at a time, address feedback before pushing new. The skill matches tone, runs the codex lineup, and decides when to fold reviewer feedback in. Queue state, hard gates, gh I/O, and the heartbeat live in `sweep` CLI. |
| argument-hint | <repo> [--check] [--push] [--list] [--ingest] [--watch] |
| allowed-tools | Read, Write, Edit, Bash, Glob |
You were invoked by the harness on one repo. Your job is the LLM-shaped part of pushing a PR: tone-matching the description, running the codex AI-likeness lineup, and handling reviewer feedback on already-open PRs. Pacing math, gh fetches, hard gates, status transitions, and the queue file all belong in sweep drip — call it.
Eleven PRs in two days gets you banned. One PR per merge cycle builds gravity. The code is the same; pacing is what makes the inbox welcoming.
| What you need | How to get it |
|---|---|
| List the queue | sweep drip list --repo … |
| Check whether it's time to push (runs every hard gate, returns the next pushable entry or a reason) | sweep drip check --repo … |
| Force-push the next queued entry | sweep drip push --repo … [--branch …] |
| Enqueue a new entry | sweep drip enqueue --repo … --branch … --issue N --test-cmd … |
| Run heartbeat (15-min check cycle) | sweep drip watch --repo … |
| Hard gates (test fail-on-master/pass-on-fix, staleness, org-gate, cooldown, dry-mode) | Run as part of sweep drip check — don't reimplement |
| Open PRs with new reviewer activity | sweep drip checkup --repo … |
If sweep drip check returns "blocked: ," read the reason and act on it — don't push past it. That's the skills-lack-determinism loop: deterministic gates return useful errors; the agent closes the loop. If a gate is mis-classifying, fix the gate, not this file.
The CLI gives you the diff and 5 recently merged PRs from the repo. Your job: write a title and body that match. Match title format, body length, level of detail. Don't write a description; inhabit the repo's voice. The chameleon rule.
The body's content follows a fixed shape — receipt format. Voice/tone still matches the repo's, but the structure shows the work:
<terse one-liner conclusion>
- <reason why thing X doesn't work, natural language>
- <reason why thing Y doesn't work, different angle>
- <finally what worked, and why>
[HG](https://github.com/kimjune01/sweep/blob/master/repo-hypotheses/<owner>__<repo>__<issue>.md)
Three rules for the bullets:
The [HG] link points at the full investigation trace in this repo's repo-hypotheses/ directory. The receipt is the curated headline; the HG is for the maintainer (or other contributors) who want to dig in. Most won't click; the existence of the link is the credibility signal.
The /investigate skill writes the HG file at ~/Documents/sweep/repo-hypotheses/<owner>__<repo>__<issue>.md before handing off to /drip. If the file is missing, omit the [HG] link rather than fake it.
After tone matching, shuffle the candidate description into 5 recently-merged PR descriptions from different contributors and ask /codex:
Here are 6 PR descriptions from a repo. One may be AI-generated. Which ones, and why? If you can't tell, say so.
Send diff + generated description + issue link to /gemini:
You are a maintainer seeing this for the first time. Would you merge it?
Five rounds max. Won't converge to zero findings — iterate until structure is sound.
sweep drip checkup reports new comments on your open PRs. For each one:
--head <user>:<branch>. You're a guest.~/.sweep/drip-queue/*.jsonl. (sweep drip enqueue / the CLI does.)gh auth status, gh pr list, gh pr view, gh issue view. (The CLI does, with the right cache and rate-limit handling.)sweep drip check. The skill doesn't replicate them.sweep drip watch or /loop 15m /drip.)max_open or max_open_per_org. (Retro params, surfaced by the CLI.)After all narration, the last printed line must be a single JSON object matching this schema:
{
"pushed": false,
"pr_url": "https://github.com/owner/repo/pull/N or null",
"outcome": "pushed | rebased | checked | noop | failed",
"reason": "short string, ≤200 chars",
"rejected": false,
"reject_reason": null
}
pushed=true iff a commit or PR actually hit the remote on this run. outcome is the verb that happened (the checked branch corresponds to --check invocations that don't push). Set rejected=true with a reject_reason when the job is undeliverable (no fork remote, stale branch, missing PR — anything that means "this msg cannot be acted on"). Rejected jobs route to ~/.sweep/inbox/rejected.jsonl for operator review.
Print the JSON as the literal last line, no trailing prose, no markdown fence.