{"slug": "the-good-ideas-were-already-there-near-the-bottom", "title": "The good ideas were already there, near the bottom", "summary": "A blind test of two Claude Code ambition skills on 9 real tasks found that neither Nick Beaird's `ambition` skill nor RoboNuggets' `be-ambitious` skill beat a plain agent with no skill on the first 6 tasks, scoring 5.64 and 4.92 overall out of 10 respectively against 5.67 for the no-skill control. The author then wrote a custom version that beat the original `ambition` skill on all 9 tasks in the final round, with GPT-6-Astra (via Codex) and Grok judging shuffled, unlabeled answers across 5 criteria; the skill and test harness are published in the Kompas/ambition GitHub repository. The test ran each combination 3 times in a fresh Claude Opus 5.5 session with no tools, using Dutch-language prompts drawn from a week of the author's work, including a matching API that matched only 54% of a customer batch and 1 in 6 jobs lost in large batches.", "body_md": "I had Nick Beaird's `ambition` skill installed in Claude Code when I came across `be-ambitious` by RoboNuggets, and I wanted to know which one was better. I'd read both. I still couldn't tell you.\n\nSo I tested them, blind, on 9 real tasks from my own week, against the same agent with no skill at all. On the first 6 tasks neither skill scored higher than the plain agent. Then I wrote my own version. In the final round it beat the original on all 9 tasks. The skill and the test harness are in [Kompas/ambition](https://github.com/Kompas/ambition), with the full results.\n\n## Where the skills come from\n\nIn October Felix Rieseberg wrote [How to be more ambitious: A tutorial for humans with models](https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/). It's worth reading. His point is that you usually ask a model for the task you already had in your head. That task is often a guess at what you want.\n\nHe describes 4 kinds of ambition. Speed is the same work, faster. Altitude aims at the outcome, so \"fix this crash\" becomes \"keep the app crash-free\". Depth means working like the best in the field, starting from the raw data, using their instruments and letting something other than the model check the result. Range brings in other fields, including an adversary, and borrows their methods.\n\nNick turned this into [`ambition`](https://github.com/Nickanbe/ambition) ([write-up](https://www.nickbeaird.com/ambition)). It gives a baseline, a ranked table of upgrades that each name an outside check, and a rewritten brief. Then it stops and you pick. RoboNuggets made [`be-ambitious`](https://github.com/robonuggets/be-ambitious), which writes an ambitious plan and then starts on the work.\n\nFelix also says a model shouldn't grade its own work. That seemed like a good rule to apply to the skills themselves.\n\n## How I tested them\n\nI took 6 requests from 1 week of work, from the chat with my business partner Tachmy and from the channel where 2 of our agents coordinate. Each came with the context an agent would have had at that point.\n\n- A missing field in a matching API, after a customer batch matched only 54%\n- Checking whether a new country works\n- 1 in 6 jobs getting lost in large batches\n- An idea for a new service\n- A YouTube video that feels too AI-made\n- 1 small question, \"Where do we still use Atlassian?\"\n\nEach skill got the same prompt, \"Be ambitious with this task\", and so did the same agent with no skill as a control. Without a control you can't tell whether a skill does anything. Every combination ran 3 times, in a fresh Claude Opus 5.5 session with no tools and none of my own settings. Here's roughly the command for 1 run.\n\n```\nclaude -p --model opus --system-prompt \"$SKILL\" --tools \"\" \\\n  --setting-sources project --strict-mcp-config --no-session-persistence \\\n  < task.txt\n```\n\nI didn't want Claude grading Claude, so the judges were GPT-6-Astra (through Codex) and Grok. They saw all answers to a task at once, shuffled and unlabeled. Each answer got a score on 5 things, and the judges named its biggest weakness.\n\n- Would the owner adopt it\n- Each upgrade has an outside check\n- It sticks to the given context\n- It makes the task deeper without making it bigger\n- A busy person can act on it right away\n\nThe tasks and the prompt were in Dutch, so the quotes below are translated.\n\n## The skills didn't beat the plain agent\n\n| Variant | Overall (1-10) | Outside check | Scope | Usable | \n|---|---|---|---|---|\n| no skill | 5.67 | 2.50 | 3.22 | 3.14 | \n| `ambition` | 5.64 | 3.56 | 2.78 | 2.72 | \n| `be-ambitious` | 4.92 | 3.19 | 2.39 | 1.69 | \n\n`ambition` did what Felix asks, and of the 3 it scored highest on outside checks. 1 run on the lost jobs even questioned the task's own assumption. Maybe the product codes Google UK doesn't know were burning through all 15 retries, and maybe that retry storm is what pushed the server into its time limit.\n\nThat idea sat in an answer of about 1,300 words. For \"Where do we still use Atlassian?\" the 2 skills wrote 550 to 800 words, with an access audit and a migration plan. The answer was 1 line. Bitbucket, for 1 old repo.\n\n## v2 changed what the agent writes down\n\nThe judges kept naming the same weaknesses. I kept Felix's method and changed only the output.\n\n| What the judges flagged | What v2 does instead | \n|---|---|\n| No answer at the top | A fixed order. The answer, what's known, up to 3 upgrades, a brief, what it needs from you | \n| Long answers | Works through Altitude, Depth and Range without printing them | \n| Invented facts, like \"the existing golden set\" | A Known block. Anything not in it is labeled \"(assumption)\" | \n| The table says you pick, the brief adopts everything | The brief holds the baseline plus upgrade #1. The rest get 1 line each | \n| Fixes before the cause is known | Measure first when the cause is unknown | \n| Targets the data says are out of reach | Targets must fit the numbers in Known | \n| 600 words for a 1-line question | Small task. Answer in line 1, under 80 words | \n\nThe skill itself went from 1,885 to 909 words.\n\nv2 won round 2 with 6.06, against 5.42 for no skill and 5.25 for `ambition`. Its scope score went to 3.92 while `ambition` stayed at 2.64. On outside checks it scored lower than `ambition`, 2.89 against 3.56. The weaknesses showed why. v2 often listed \"me, today\" as the check, which is the agent grading itself again.\n\n## Then v2 lost on new tasks\n\nI wrote v2 from the mistakes on the same 6 tasks it then won on. That's sitting a test when you already know the questions. So I added 3 tasks I hadn't looked at while writing it. A nightly spike of useless requests, a monitoring service exposed to the internet, and how to get our new site ranking in Google.\n\nOn those, `ambition` scored 6.44 and v2 6.06. No skill got 4.78. v2 was shorter and easier to use. `ambition` had the better ideas.\n\nWithout those 3 tasks I would have published v2 as the winner.\n\n## v3 added 2 rules\n\nThe first rule is that the check is never the agent. \"I'll review the logs\" doesn't count. \"The error count in Grafana for that batch\" does, and the skill asks who reads it, when, and what counts as a pass.\n\nThe second is that a requested feature gets built. Measuring first is for a problem with an unknown cause. When someone asks for a feature, the agent delivers it and the upgrades build on it.\n\nThe final blind round had every variant on all 9 tasks.\n\n| Variant | 6 original tasks | 3 new tasks | Median length | \n|---|---|---|---|\n| v3 | 6.28 | 6.44 | 487 words | \n| v2 | 6.08 | 6.06 | 494 words | \n| `ambition` | 4.83 | 6.00 | 573 words | \n| no skill | 5.06 | 4.67 | 858 words | \n\nv3 scored higher than `ambition` on all 9 tasks, by 1.1 points on average (95% interval 0.7 to 1.6). Both judges, separately, ranked v3, v2 and `ambition` in that order.\n\nTake the Atlassian question. `ambition` opens with a header and a baseline section that does mention Bitbucket, and then roughly 500 more words follow. v3 answered in 76 to 87 words. In the first run, line 1 says that as far as it knows only Bitbucket is in use, for 1 repo. Then 1 upgrade, because the real question is probably whether the subscription can go, so put the billing lines next to the last activity per product.\n\nAll 3 v3 runs on the lost jobs opened with the same advice. Find out where those jobs fail before fixing anything.\n\n## What this doesn't show\n\nThe judges are models. They judged blind and come from other labs, but I haven't ranked the answers myself yet. It's also 1 company, 1 week and 1 model, with 9 tasks and 3 runs each.\n\nThe agents had no tools, so I measured plans, not what happened when someone ran them. That hurts `be-ambitious` most, since it's built to start working right away. Nick's [own eval](https://github.com/Nickanbe/ambition) does run the brief against test fixtures, and mine doesn't.\n\nI also wrote both v3 and the scoring questions. Those questions reward a short answer at the top, which is what v3 was built to give. The 3 new tasks protect against fitting to the tasks. They don't protect against fitting to the rubric.\n\n2 results don't flatter v3. On the new tasks `ambition` scored highest on value and was picked as the best single answer most often. v3 wins on average because it's easier to use, and I'm not claiming it thinks better. v3 also misses its own length limit. The skill says under 350 words, and its answers ran around 490.\n\n## Try it on your own tasks\n\nThe harness in [Kompas/ambition](https://github.com/Kompas/ambition) runs any set of skills against a no-skill control, and Codex and Grok judge the answers blind. The [full results](https://github.com/Kompas/ambition/blob/main/eval/RESULTS.md) are there too. What wins on my week may not win on yours.\n\n2 things made the difference for me. The control, because without it I wouldn't have seen that the original skills added nothing on average on my first 6 tasks. And the 3 tasks I held back, because they're the reason I didn't publish v2 as the winner.\n\nThe method is [Felix Rieseberg's](https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/). The first skill is [Nick Beaird's](https://www.nickbeaird.com/ambition), and it was my starting point. The second is by [RoboNuggets](https://github.com/robonuggets/be-ambitious).", "url": "https://wpnews.pro/news/the-good-ideas-were-already-there-near-the-bottom", "canonical_source": "https://blog.hagebols.com/posts/ambition-skill/", "published_at": "2026-11-08 00:00:00+00:00", "updated_at": "2026-11-13 10:19:40.438360+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "large-language-models", "ai-research"], "entities": ["Nick Beaird", "RoboNuggets", "Felix Rieseberg", "Claude Code", "Claude Opus 5.5", "GPT-6-Astra", "Grok", "Kompas/ambition"], "also_reported_by": [], "takeaway": "Neither the `ambition` (5.64) nor `be-ambitious` (4.92) skill beat the no-skill control (5.67) on the first 6 tasks, while the author's rewritten version beat the original `ambition` skill on all 9 tasks.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/the-good-ideas-were-already-there-near-the-bottom", "markdown": "https://wpnews.pro/news/the-good-ideas-were-already-there-near-the-bottom.md", "text": "https://wpnews.pro/news/the-good-ideas-were-already-there-near-the-bottom.txt", "jsonld": "https://wpnews.pro/news/the-good-ideas-were-already-there-near-the-bottom.jsonld"}}