zaidmukaddam

zaidmukaddam/miniscira

Instructions file CodexOpenCode

Instructions for zaidmukaddam/miniscira, covering project conventions, toolchain, non-obvious invariants, motion tokens and this is not the next.js you know.

zaidmukaddam/skills

Skill Claude CodeCodex

Write commit messages as a single line with a conventional type prefix, the way they actually get written here. The default shape is type: what changed on one line, lowercase after the colon, no trailing period, no body. Pick the type from feat, fix, refactor, chore, docs, and style. When one commit genuinely does…

zaidmukaddam/skills

Skill Claude CodeCodex

Draft a post for X against what the published For You ranking code actually rewards, rather than against folk wisdom about engagement. Takes a mode argument. Use launch to announce something new, update to report progress on something an audience already knows about, and banger when the goal is maximum reach. The…

zaidmukaddam/skills

Skill Claude CodeCodex

Never edit tracked files by running a script. Banned are sed -i, perl -pi, python or node or bun one-liners and heredocs that write files, awk rewrites, and shell redirects onto source, because each one lands a change with no reviewable diff, fails silently when its pattern misses, and reports success either way. Use…

zaidmukaddam/skills

Skill Claude CodeCodex

Shape a change so it can actually be reviewed, not just merged. Decide before writing code whether the work is one pull request or a stack of them, keep the title and body describing the diff as it exists right now, not the plan you started with, group commits by concern in dependency order, give every review comment…

zaidmukaddam/skills

Skill Claude CodeCodex

Rebuild a decision from what must be true instead of from precedent. Recover the real objective, label every assumption, compute the theoretical floor and compare the current system against it, delete before optimizing, then propose the cheapest test that would settle the remaining doubt. Use this whenever the user is…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply design engineer judgment, the role that owns the shipped result rather than a handoff: decide the intent before the pixels, name the defaults you are rejecting, build hierarchy from space and weight before size, treat copy and accessibility as part of the interface, ask whether something should animate before…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply staff engineer judgment, where the unit of work is the decision, not the diff: treat complexity as the thing being minimized, design the interface twice before building once, put a seam only where something actually varies, build a tight feedback loop before building a theory, state claims so they can come back…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Dillon Mulroy's reasoning about building systems that survive contact with production: plan at the call-stack and type level before any code exists, design for the person debugging this at 2am during an incident, parse data once at the edge and refuse to let it widen deeper in, make failure modes explicit rather…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Dane Knecht's platform engineering judgment: pick the cheapest compute primitive that fits and escalate only on demonstrated need, reason from traffic composition and physical limits rather than benchmark scores, replace the substrate while preserving the contract, turn a proprietary advantage into an open…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Elon Musk's engineering reasoning: find the actual limiting factor and solve only that, reduce a system to the one scalar that dominates it and argue in that unit, judge a design by what would make it pointless rather than by current specifications, concede the axis you lose and name the axes you win, gate…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Garry Tan's founder-first judgment: judge by outcomes rather than inputs, ask who is funded by the problem continuing, use cohort retention as the one readout that tells you whether the thing is good, fund people rather than ideas, keep enough runway that hard cuts happen early, and build the replacement instead…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Andrej Karpathy's reasoning about machine intelligence: predict capability from what the supervision signal can verify rather than from benchmark scores, build the smallest artifact that contains the whole mechanism, do napkin math two ways and attach an error bar, name where your own analogy breaks, give an…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Lee Robinson's approach: build the real artifact and let it carry the argument, run comparisons on a deliberately unfriendly case with the exact prompt published, reach for the lowest abstraction that works, distrust any single aggregate number and name the confounders, teach concepts in dependency order, and…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Naval Ravikant's reasoning: treat truth as the terminal goal rather than a means, weigh feedback by whether it came from reality or from an audience, redefine the key term until the problem dissolves, prefer explanations that are hard to vary, and check what wanting the outcome is costing you. Use this when…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Guillermo Rauch's engineering judgment: count the round trips and delete one, put the compute next to the data, never show a spinner where a skeleton or an optimistic update belongs, migrate one route at a time while measuring field data, design the limits in from the start, and attach a measured number to every…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Sam Altman's reasoning: commit to a dated goal with the failure probability stated out loud, argue the asymmetry between over- and under-provisioning the scarce resource and publish the rationing order, ship to real users and let their feedback set the roadmap, publish the standing rule instead of relitigating…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply shadcn's approach to distributing and designing code: ship open code the consumer owns rather than a closed dependency, give great defaults with an escape hatch that goes all the way to zero, push variance into the install step instead of the shared source, treat error messages and schemas as an agent-facing…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply thdxr's reasoning about building developer tools and the economics under them: price the thing people actually consume rather than the sticker rate, own the primitive when nothing on the shelf meets the bar, treat open source as coverage your core team cannot buy, define the interfaces before handing anything to…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply Theo's reasoning about working with generated code and evaluating the tools that produce it: stop asking how much code you should read and start asking how much you should generate, spend cheap disposable code on verifying the code that matters, convert every repeated correction into a lint rule or a CI step so…

zaidmukaddam/skills

Skill Claude CodeCodex

Apply the engineering standards used in these codebases: match a named reference product instead of inventing a look, verify in the running app rather than the diff, delete features instead of complicating them, fix the mechanism rather than the symptom, and treat AI-slop design and unverified claims as defects. Use…