- Home
- Skills
- Research & Analysis
- Systems Paper Writing: OSDI to NSDI
More screenshots
Works with the AI Agents you already use
Systems Paper Writing: OSDI to NSDI
Structure a systems paper as an argument, not a report: one thesis sentence, sections mapped to the reviewer's three questions, evaluation built as proof, and a claim-evidence audit before you draft.
$5
Systems Paper Writing: OSDI to NSDI
Example session with this skill installed
We built an adaptive caching layer for a key-value store and we have benchmark results, but our OSDI draft reads like a lab report — chapters in the order we built things — and our advisor says reviewers will shrug. Restructure it.
- Read your context and instructions
- Compiled the systems paper writing
RESTRUCTURE PLAN — produced by applying the skill to your draft.
Step 1, thesis sentence. Your draft doesn't have one; your results support this: "Admission-aware caching beats LRU and FIFO admission for read-heavy key-value workloads with skewed access on a single node." Notice what it forces: Y (read-heavy, skewed) and Z (single node) are now explicit. Your multi-node claim in chapter 6 has no evidence behind it — it gets scoped to future work, not asserted.
Step 2, remap the sections. Your chapters 1–3 (motivation, what we built, how it works) answer no reviewer question in order. New map: Introduction answers "why should I care" — open with the cache-miss cost in your production trace (the 38% miss-rate number buried in your chapter 5 belongs in paragraph one), then name the gap in LRU and FIFO admission in one sentence each, with evidence. Design answers "is it real and is it sound" — your admission policy, plus the two alternatives you tried and rejected (you mentioned them in a footnote; promote them, a choice without alternatives reads as unexamined). State the implementation plainly: language, lines of code, what's real.
Step 3, evaluation as proof. Reorder from chronology to argument: end-to-end vs. LRU and FIFO on the skewed trace first (that's the thesis), then the ablation removing admission-awareness while keeping everything else — that's what turns "our cache is fast" into "it's fast because of the admission policy." Every conclusion gets stated three times: hypothesis before, result after, evidence in the caption.
Step 4, the audit, started for you: Claim "lower miss rate under skew" → end-to-end experiment → proved. Claim "win comes from admission, not cache size" → ablation → proved. Claim "holds under write-heavy load" → no experiment exists → owed (run it) or cut. Claim "multi-node" → cut, scoped.
Abstract gets written last, five sentences. Your current one was written first — that's why it promises the multi-node story the paper can't prove.
Connects securely to your tools. The creator never sees your data.
About this skill
Most rejected systems papers don't fail on the system — they fail on the argument. The draft describes what was built, in build order, and asks the reviewer to work out why it matters. Reviewers under time pressure don't do that work; they shrug, and the paper dies. This skill restructures the writing around the reader who actually decides: a skeptical committee member with three questions — why should I care, is it real and sound, do the numbers prove the thesis. It starts with the thesis sentence ("X is better for applications Y running in environment Z") and treats it as a contract every section serves. The evaluation is designed backward from that sentence as a proof strategy: end-to-end comparisons against baselines reviewers respect, plus ablations that attribute the win to your specific choices, with every conclusion stated three times (hypothesis, result, figure caption) so the claim survives skimming. Before results prose gets drafted, you build a claim–evidence audit: every claim mapped to the experiment that proves it, marked proved, owed, or cut — so nothing gets asserted that the data doesn't support, and scoped claims replace invented ones. Related work is grouped by approach and differentiated explicitly, with comparison tables that admit where you lose. The integrity contract is non-negotiable and stated in full: no fabricated observations, no synthetic traces passed off as real, no unverified citations. The skill ships a fill-in paper skeleton, the audit table, a worked storage-system outline with its audit filled in, and the three adversarial revision reads — skeptic, related-work expert, and the tired reviewer at 11 p.m. — that catch what the committee would.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
- It structures arguments; it cannot rescue a paper whose evaluation genuinely doesn't support any thesis — the audit will surface that as owed or cut claims, which is the honest outcome, not a fix.
- Venue rules (page limits, disclosure policies, formatting) change yearly; the skill tells you to verify against the current call but doesn't track them for you.
- The worked examples come from storage and caching systems; the structure transfers across systems areas, but area-specific evaluation norms (e.g., for networking or security papers) still need your judgment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected