Five rules for evaluating agent skills before they bloat your context window
Comments on “Five rules for evaluating agent skills before they bloat your context window”
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
A Google Cloud developer-relations post lays out a five-rule playbook for checking whether agent skills (SKILL.md files) justify their context cost. The rules cover catching trigger collisions before running containers, running evaluations in isolated sandboxes rather than on laptops, comparing paired with/without-skill runs over repeated trials, keeping correctness separate from…
Why it ranked
No. 135 of 139 with a score of 0.55. The panel rated it highest on practical relevance (0.82) and technical depth (0.76), lowest on significance (0.50). The judges disagreed most on promotional (spread 0.55), so the story is marked contested. Placed No. 135, No. 136 by score alone: the 57th AI & machine learning story of the day (−0.080). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Practical, detailed guidance on evaluating agent skills with concrete eval-methodology lessons, but partly vendor-promotional and mostly tied to one company's plugin and tooling.
- AI judgeJudge 3gpt-5.6-luna
- Strongly relevant engineering guidance on evaluating agent skills, sandboxing, ablation, reward hacking, and security. It offers concrete methods and claims, but evidence is mostly internal and the closing plugin promotion is substantial.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.55total under rubric v2.3Contestedjudges differ by 0.44 on new information, 0.55 on promotional
- Significance0.50
- Reader interest0.72
- Audience breadth0.61
- Practical relevance0.82
- Substantiated0.62
- Technical depth0.76
- New information0.67
- Counts against: Promotional0.46
- Counts against: Overclaim0.28
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.57 | 0.30 | 0.62 | 0.50 | 0.32 | +0.24 | +0.119 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.75 | 0.60 | 0.82 | 0.72 | 0.22 | +0.26 | +0.188 |
| Audience breadthHow many people who build technology does it concern? | 0.65 | 0.50 | 0.68 | 0.61 | 0.18 | +0.13 | +0.079 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.92 | 0.70 | 0.84 | 0.82 | 0.22 | +0.22 | +0.180 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.66 | 0.50 | 0.70 | 0.62 | 0.20 | +0.05 | +0.031 |
| Technical depthHow technically substantive is it? | 0.92 | 0.60 | 0.76 | 0.76 | 0.32 | +0.05 | +0.038 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.84 | 0.40 | 0.78 | 0.67 | 0.44 , above the contested threshold | +0.05 | +0.034 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.17 | 0.50 | 0.72 | 0.46 | 0.55 , above the contested threshold | −0.19 | −0.088 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.25 | 0.30 | 0.28 | 0.28 | 0.05 | −0.11 | −0.030 |
| Sum of contributions | +0.552 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.551 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-05/135/"><img src="https://rubric.hrih.si/badge/2026-11-05/135.svg" alt="Ranked #135 on Rubric, 5 Nov 2026" width="245" height="28"></a>Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 3 pts, not on front page
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.