build-without-fear
Skill1 starsChanged 2 months agoMITView on GitHub
Design a complete subsystem when requested or when recurring defects require explicit state, ownership, and recovery.
In this repository16 other files it gives its agents.
- adversarial-refinement-loopSkill
- atomic-commitsSkill
- audio-verificationSkill
- comfyui-assetsSkill
- diagnosing-bugsSkill
- finish-the-listSkill
- frontend-designSkill
- headless-terminal-workflowSkill
- parallelize-independent-workSkill
- prefer-second-monitorSkill
- require-external-communication-approvalSkill
- state-of-the-art-firstSkill
- try-it-firstSkill
- visual-verificationSkill
- web-design-guidelinesSkill
- writing-for-agentsSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 12 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈12
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 28
When evidence shows several bugs share a boundary, repair its ownership or transitions coherently.
- Distinctive: line 42
Keep business rules separate from framework adapters.
- Distinctive: line 48
Prefer exhaustive discriminated unions and total switches.
- Distinctive: line 49
Prefer named policies and tables over unexplained constants.
- Distinctive: line 50
Prefer separate preparation, commit and cleanup phases.
- Distinctive: line 51
Prefer idempotent operations and monotonic generations over timing guesses.
- Distinctive: line 52
Prefer bounded queues, caches and instance pools with documented eviction.
- Very specific: line 53
Prefer a full scene/snapshot/model description over scattered DOM/CSS patches.
- Distinctive: line 54
Prefer comments that explain ordering, ownership and failure modes.
- Distinctive: line 71
update only transition-owned presentation state.
- Distinctive: line 76
Never use an arbitrary delay as the primary correctness mechanism.
- Distinctive: line 79
Do not accept a handful of screenshots or one Playwright gesture as proof of a temporal fix.
- Distinctive: line 88
Before stopping, reread the request and account for every item.
≈ “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
Build Without Fear
Build the system required by the requested behavior. Base architectural depth on observed failure modes and supported product states. Code size is neither a target nor a constraint. Keep routine component changes focused.
Start from the state space
Identify the dimensions relevant to this subsystem:
- lifecycle states, transitions, interruptions, retries, cancellation and teardown;
- success, loading, empty, stale, partial, error, fallback and recovery states;
- input methods, timing orders, re-entrancy, rapid repetition and concurrency;
- platform, renderer, device, accessibility and reduced-capability variants;
- persisted, cached, derived, live and presentation state;
- ownership boundaries and which subsystem is authoritative at each instant.
Represent relevant invariants through types, state machines, or contracts as useful. Avoid boolean piles whose combinations are accidental.
Prefer a subsystem over patches
When evidence shows several bugs share a boundary, repair its ownership or transitions coherently. Replace it only when a focused correction is inadequate. Choose the modules the subsystem needs from concerns such as:
- domain types and invariants;
- state machine or coordinator;
- preparation and preload stages;
- runtime execution;
- interruption, retry and recovery;
- adapters for frameworks or platforms;
- diagnostics and inspectable status;
- cleanup and resource ownership;
- compatibility and migration paths.
Keep business rules separate from framework adapters. Expose the operation states callers actually need to distinguish, such as queued, failed, cancelled, or completed.
Write explicit code
- Prefer exhaustive discriminated unions and total switches.
- Prefer named policies and tables over unexplained constants.
- Prefer separate preparation, commit and cleanup phases.
- Prefer idempotent operations and monotonic generations over timing guesses.
- Prefer bounded queues, caches and instance pools with documented eviction.
- Prefer a full scene/snapshot/model description over scattered DOM/CSS patches.
- Prefer comments that explain ordering, ownership and failure modes.
- Preserve extension points for plausible product variants already visible in the codebase; do not invent unrelated features merely to increase size.
Implementation size should follow the actual state space. Avoid filler, unused abstractions, and speculative architecture. Every additional line should cover a real state, make ownership explicit, improve craft, or make failure recoverable.
Handle temporal bugs as transactions
For animation, audio, rendering and UI transitions, treat each operation as a transaction:
- prepare every resource and destination state;
- validate readiness and choose a declared fallback;
- atomically enter the transition;
- update only transition-owned presentation state;
- atomically commit the destination;
- release obsolete resources after the first committed presentation opportunity;
- ignore late work using generation tokens or abort signals.
Never use an arbitrary delay as the primary correctness mechanism. If a resource can be late, represent that fact and decide what the product shows while it is late.
Do not accept a handful of screenshots or one Playwright gesture as proof of a temporal fix. Those probes can alias past a short fault and headless scheduling can change the behavior. Use them to inspect named phases, but combine them with code-level ordering/resource-readiness analysis and repeated normal, rapid and interrupted interaction. A user's repeatable observation in the real app outweighs an automated non-reproduction.
Finish the implementation
Before stopping, reread the request and account for every item. Report any possibility deliberately excluded and why. Run only the level of verification the user permits; if verification is restricted, do not substitute stronger claims than the evidence supports.
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.

