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 skills add nikships/skills-registry --skill pr_alignment_loopgit clone --depth 1 https://github.com/nikships/skills-registryWrote 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/nikships/skills-registry/pr_alignment_loop)<a href="https://agentmods.dev/skills/nikships/skills-registry/pr_alignment_loop"><img src="https://agentmods.dev/badge/skills/nikships/skills-registry/pr_alignment_loop.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 165 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
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 | $0.00122 | $0.02068 |
| Opus 5 | $0.00061 | $0.01034 |
| Sonnet 5 | $0.00024 | $0.00414 |
| Haiku 4.5 | $0.00012 | $0.00207 |
Grade A, and why
pr-alignment-loop 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 — 166 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PR Alignment Loop
You orchestrate two read-only reviewer droids over a real PR until it is merge-ready. You own the edits, the tiebreaks, and the stopping decision. The reviewers only critique.
The two droids:
reviewer-robustness— biased toward correctness, edge cases, security, failure modes.reviewer-minimalist— biased toward simplicity, deletion, anti-speculative-complexity.
They are intentionally biased in opposite directions. The negotiated middle is the goal — not the union of their wishlists.
Operating principles
- You are the implementer. Reviewers never edit code. You synthesize their feedback and edit.
- Bias to ship. Default is APPROVE once no
HIGH-severity issue remains. Do not chase polish. - Anti-bloat is co-equal with anti-bug. Adding code has a cost. Do not action
MED/LOWrobustness findings if the minimalist disagrees and there is no concrete failure path. - Hard 3-round cap. No exceptions.
- Log every conflict resolution. When the two reviewers disagree, write down which side you took and why, in one sentence, in the final report.
Inputs
The user will give you one of:
- A PR URL (
https://github.com/<owner>/<repo>/pull/<n>) - A PR number with repo context
- A branch name with an implicit "review this branch's diff against base"
If ambiguous, ask once for the PR identifier. Do not start the loop without it.
Setup (before round 0)
- Use
ghCLI for everything PR-related. NeverFetchUrlGitHub URLs (per global rules).gh pr view <id> --json title,body,baseRefName,headRefName,files,additions,deletionsgh pr checkout <id>gh pr diff <id>— capture the unified diff
- Identify the project's verification commands by inspecting:
package.jsonscripts → prefertest,lint,typecheck,buildMakefile→ look fortest,checkpyproject.toml/pytest.ini/tox.iniCargo.toml,go.mod, etc. Record the commands you will run after each edit pass. If you cannot find any, note that — you'll skip the verification step but flag it in the final report.
- Note the PR scope from the description. Anything the PR intentionally doesn't do is out of scope for reviewer findings.
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 · 166 lines · 122 tokens per session scan A 55278f4684f7
pr-alignment-loop is a skill published in the GitHub repository nikships/skills-registry (17 stars, last pushed 15d ago), licensed Apache-2.0. It adds 122 tokens to every session and 2,068 once invoked, about $0.0006 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
deep-review
Evidence-based deep code review rubric — severity + confidence findings across maintainability, correctness, security, performance, and testing. Push for ambitious structural simplification (code judo), not just local cleanup. Escalate hands-on triage to the code-reviewer agent.
reviewing-update-sets
Reviewing, analyzing, and comparing ServiceNow update sets before promotion. Shows changes, risks, dependencies, and conflicts. Use when the user mentions update sets, customizations, promotion, code review, change tracking, pre-deployment review, sysupdatexml, customer updates, "what changed in this update set," or…
gh-address-comments
Triage and address open GitHub PR review and conversation comments using the gh CLI. Use when the user wants to "address PR comments", "resolve review threads", or "respond to reviewers" on the current branch's pull request.
deslop
Remove AI-generated code slop and clean up code style.
review-all
Use when the user asks for a deep review, full review, comprehensive review, production readiness assessment, full audit, multi-domain audit, "security and reliability and code review", or "review everything". Also use when the user explicitly requests performance review alongside the comprehensive request (e.g.…
active-review
Use when the user wants to prepare for a manual PR code review and asks for help targeting it — a terse PR summary, ranked files to read first, and paste-ready inline comment drafts with GitHub deep-links. Triggers: "active review this PR", "walk me through this PR", "help me review this", "give me inline comments to…