awb-research
Skill2 starsChanged 6 months agoMITView on GitHub
WHAT: understand a problem before building — scope it, read the actual code, compare at least two approaches, then recommend one with reasons. USE WHEN: a non-trivial change where the approach isn't obvious ("how should we...", "what's the best way to...", "look into X before we build it", "which option..."). DO NOT TRIGGER: a vague *request* that needs clarifying first (that's prompt-refiner); executing an already-chosen approach (that's a plan-then-code skill); a known one-line fix.
In this repository21 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- awb-config-guardSkill
- awb-cookSkill
- awb-dead-code-auditSkill
- awb-debugSkill
- awb-external-refSkill
- awb-handoverSkill
- awb-install-and-verifySkill
- awb-lessons-captureSkill
- awb-optimizeSkill
- awb-output-guardSkill
- awb-plan-then-codeSkill
- awb-reviewSkill
- awb-session-closeSkill
- awb-stress-testSkill
- awb-tddSkill
- awb-uninstallSkill
- awb-using-skillsSkill
- awb-verify-docsSkill
- prompt-refinerSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
Zoom out first: Before diving into one file, go up one layer: what module owns this, who calls it, where does the data come from and go?
- Distinctive: line 27
Explore — read the real code: Read the relevant code and configuration *as it is now*, not as you remember or assume it.
- Distinctive: line 30
HARD GATE: at least two approaches, in a table: Do not recommend until you have compared ≥ 2 concrete options across the dimensions that …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Research before building
Announce on activation: "Using awb-research — I'll read the code and compare options before recommending."
The expensive mistake is committing to an approach you picked before you understood the terrain. This skill forces understand → compare → decide, on evidence, before any plan or code.
Process
- Zoom out first. Before diving into one file, go up one layer: what module owns this, who calls it, where does the data come from and go? A quick map of the neighbourhood stops you solving the wrong problem in the right file. Don't deep-read yet — orient.
- Scope. State the question in one sentence and the constraints that actually bind (perf, compatibility, deadline, the existing stack). An approach that ignores a real constraint is noise.
- Explore — read the real code. Read the relevant code and configuration as it is now, not as you remember or assume it. Note what exists already (a helper, a pattern, a prior decision) so you don't reinvent or contradict it.
HARD GATE: at least two approaches, in a table. Do not recommend until you have compared ≥ 2 concrete options across the dimensions that matter (effort, risk, fit with existing code, reversibility). One option is not a comparison — it's a foregone conclusion dressed up.
Approach Effort Risk Fits existing code? Reversible? A — … … … … … B — … … … … … - Recommend one, with the reason and the trade-off you accept. Name the option, why it wins given the constraints, and what you're giving up by not picking the runner-up. Then hand off to a plan/implement skill — research stops at the decision, it doesn't build.
Banned behaviours
- Proposing an approach before reading the code it touches ("I'd probably use X") — that's a guess wearing a recommendation's clothes.
- A "comparison" with one real option and a strawman.
- Recommending without naming the trade-off you're accepting.
Anti-rationalization
| You'll think | Reality |
| "I already know how this codebase works" | Memory drifts; the code is the source of truth. A 5-minute read beats a confident wrong assumption. |
| "There's obviously only one way" | If it's truly obvious, the comparison takes two minutes and confirms it. If it's not, you just dodged a rebuild. |
| "Comparing is slower than just building" | Building the wrong approach and reversing it is the slow path. The table is the cheap one. |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

