Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →jpvarbed/apply-paper
Turn a research paper's finding into a concrete change to our skills/agents — distill the claim, map it to where it applies, design the change, and prove it helps.
Passed 19 of 22 test cases with the skill, and 9 of 22 without it.
and 11 more in the full report
10 fixed9 passed either way3 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
Applying 'ReAct' reasoning loops to `agents/orchestrator.ts` requires a complex, multi-step migration across four sprint tasks. Beyond updating `docs/papers.md`, what tracking action should be taken?
| Checked | Without skill | With skill |
|---|---|---|
| The response instructs filing or refreshing a Linear JAS issue | Missed | Passed |
The instructions your model reads when the skill loads.
Closes the loop on docs/papers.md: that file captures papers; this skill applies them.
A paper only earns its keep when its mechanism changes how a skill or agent behaves. The output
is a change (or a tight change plan), not a summary.
In 3-4 sentences: what does the paper actually show, and by what mechanism? Separate the finding (often robust) from the framing (often oversold). If you can't name a mechanism, you can't apply it — stop and say so.
Name the concrete target: which skill / agent / instruction-layer / flow does the mechanism
touch? Cite the file. If nothing in our setup is affected, say "no application" and log that in
docs/papers.md — that's a valid, honest outcome.
The smallest edit that captures the mechanism: a new check, a new mode/step, a tightened contract, a deleted footgun. Prefer scoped edits over broad rewrites (HarnessFix's own lesson). Write it as a diff plan: file → what changes → why it captures the mechanism.
State the change's done-condition as a check, then show before/after:
Red-team with adversarial-review: does the change actually capture the paper's mechanism, or
just cite it? Could it be satisfied without the intended effect (cargo-culting the paper)?
Update the paper's entry in docs/papers.md with → applied: <what changed> (<file>), so the
list shows captured-vs-shipped at a glance. File/refresh a Linear JAS issue if it's multi-step.
PAPER: <title + url>
CLAIM: <mechanism in one sentence>
test cases passed, vs 9 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.