Sina review

Scientist agent mode for AI coding agents — hypothesis-driven, evidence-based, uncertainty-aware.

Install

npx -y skills add srahnama/Sina --skill sina-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Review focused on unsupported claims, missing evidence, and scientific discipline. Finds what lacks proof: untested fixes, correlation treated as causation, fabricated or vague citations, fixes without repro, shotgun debugging. One line per finding: location, problem, required evidence. Use when the user says "review for evidence", "what's unproven", "sina review", "check the reasoning", or invokes /sina-review. Complements correctness-focused review; this one only hunts unsupported claims.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.4 KB, as published. Nobody here has run it

Review diffs, reports, and reasoning for missing evidence. One line per finding: location, what's wrong, what evidence is required.

Format

L<line>: <tag> <problem>. <required evidence>., or <file>:L<line>: ... for multi-file reviews.

Tags:

untested: fix or claim with no repro or experiment recorded. Run the failing case; show before/after.
guess: "probably" / "likely" with no hypothesis or test. State H, define falsifier, run minimal check.
causal: correlation or mechanism stated as proven cause. Name confounders or design gap.
citation: source missing, vague, or unverifiable. Cite verified source or label [HYPOTHESIS]/[UNKNOWN].
shotgun: multiple unrelated edits masking diagnosis. One hypothesis → one change → observe.
premature: "fixed" / "done" without verification experiment. Rerun case; check regression.
overclaim: confidence exceeds evidence (high from one log line). Downgrade or add confirmatory test.

Examples

❌ "This looks like it might be a race condition, consider adding a mutex."

✅ L44: guess. Race assumed, no ordering evidence. Add timestamps on both paths; falsify if events are sequential.

✅ auth.py:L88: untested. Guard added, no repro steps. Show 401 before and 200 after with same credentials.

✅ report.md:L12: citation. "Studies show X" with no source. Verify primary literature or mark [UNKNOWN].

✅ diff: shotgun. 6 files changed for one bug. Isolate H1 in one file first.

Scoring

End with: evidence gaps: <N>. If reasoning is tight and verified, say Evidence-backed. Ship. and stop.

Boundaries

Scope: evidence and reasoning discipline only. Style, naming, and formatting are out of scope unless they hide an unsupported claim. Does not apply fixes, only lists gaps. "stop sina-review" / "normal mode": revert to normal review style.