nyk (@nykdotdev) header
nyk (@nykdotdev) profile picture

nyk

@nykdotdev

exploring intelligence, markets & the systems shaping both · building @chartrunner_xyz · prev founded @builderzdotdev · @rpcedge · member @SuperteamDE

 remote
builderz.dev
Joined September 2017
  • Tweets19.4K
  • Following946
  • Followers9.8K
before you call a coding agent setup production, map six gates: 1. authority 2. capability admission 3. isolation 4. evidence 5. release approval 6. recovery a green test run can still fail evidence, boundary, approval and rollback. the production agent systems course works through each gate with scenario labs. nyk.dev/products/production-…
Screenshot of 'The product' section on the Production Agent Systems course page at nyk.dev. Heading about the complete production method without scheduled sessions. Checklist: Start Here diagnostic, six module PDFs with eighteen lessons, six scenario labs with annotated solutions, capstone workbook and field guide, private online reader, editable Markdown source, Agent OS Kit v1.1.0 included.

ALT Screenshot of 'The product' section on the Production Agent Systems course page at nyk.dev. Heading about the complete production method without scheduled sessions. Checklist: Start Here diagnostic, six module PDFs with eighteen lessons, six scenario labs with annotated solutions, capstone workbook and field guide, private online reader, editable Markdown source, Agent OS Kit v1.1.0 included.

5
524
closed the tab that ran the dev server and the flags went with it. --port 3000 or 3001? devports rebuilds the command from the running process, shows it first, then restarts it in a fresh terminal window. env vars don't come along, so exports belong in project config.
Screenshot of the Restart section on devports.xyz. Left: headline 'Restart without hunting for the tab.' and a line saying DevPorts rebuilds the command that started the server, shows it, then stops the server and runs it in a fresh Terminal window. Right: dialog 'Restart storefront?' showing cd ~/code/storefront && pnpm dev --port 3000 with Cancel and Restart buttons. Sample values are the site's.

ALT Screenshot of the Restart section on devports.xyz. Left: headline 'Restart without hunting for the tab.' and a line saying DevPorts rebuilds the command that started the server, shows it, then stops the server and runs it in a fresh Terminal window. Right: dialog 'Restart storefront?' showing cd ~/code/storefront && pnpm dev --port 3000 with Cancel and Restart buttons. Sample values are the site's.

311
generic instruction files do not know how your repo builds, tests, or recovers. audit one next.js repo, preview the patch, apply only with --yes, verify with the repo's own checks, keep a rollback receipt. nyk.dev/products/harness-rea…
Screenshot from the Harness Ready page, self-serve workflow section. Headline: Audit first. Approve the patch. Keep the receipt. Three steps: audit the repository (read-only), preview the exact plan before write consent, apply then verify with a receipt-scoped rollback path.

ALT Screenshot from the Harness Ready page, self-serve workflow section. Headline: Audit first. Approve the patch. Keep the receipt. Three steps: audit the repository (read-only), preview the exact plan before write consent, apply then verify with a receipt-scoped rollback path.

1
3
363
a draft can look fine until something names the line. copylint lists every finding with its rule and excerpt, then exits 1 at score 40. same number in the browser demo and in the action.
Screenshot of the Copylint demo on copylint.xyz with sample launch-email.md. Left: draft full of stock phrases. Right: FAIL 100/100 against floor 40, exit 1, critical/major/minor counts, each finding showing line, rule, and excerpt (game-changer, evolving landscape, let's dive in). Caption notes the scan runs in the browser and sends nothing.

ALT Screenshot of the Copylint demo on copylint.xyz with sample launch-email.md. Left: draft full of stock phrases. Right: FAIL 100/100 against floor 40, exit 1, critical/major/minor counts, each finding showing line, rule, and excerpt (game-changer, evolving landscape, let's dive in). Caption notes the scan runs in the browser and sends nothing.

1
2
377
third-party skills arrive as a zip and a promise. inventory the bundle, flag shell network filesystem and secrets signals, then record who decided and why before anything installs. nyk.dev/products/skill-vetti…
Screenshot from the Skill Vetting Pack page, "The product" section. Six checks: dependency-free local CLI for inventory and comparison; content digests and review pointers for shell, network, filesystem, secrets, dynamic execution, and installation; four-tier risk model; editable review record with decision, evidence, owner, conditions, and re-vetting triggers; large-bundle triage playbooks; worked example and offline regression tests.

ALT Screenshot from the Skill Vetting Pack page, "The product" section. Six checks: dependency-free local CLI for inventory and comparison; content digests and review pointers for shell, network, filesystem, secrets, dynamic execution, and installation; four-tier risk model; editable review record with decision, evidence, owner, conditions, and re-vetting triggers; large-bundle triage playbooks; worked example and offline regression tests.

3
1
385
watchers and workers do not listen on a port, so lsof never shows the ones your agents left behind. orphaned processes keep eating memory until someone names them. devports puts them under background with the command and which agent started them.
Screenshot from the DevPorts site, Background section. Left: menu bar app listing Background with one orphaned watcher and two Claude Code tagged workers. Right: "Find what your agents left running" — watchers and orphaned processes that do not listen on a port, listed with command, project, memory, and which agent started them.

ALT Screenshot from the DevPorts site, Background section. Left: menu bar app listing Background with one orphaned watcher and two Claude Code tagged workers. Right: "Find what your agents left running" — watchers and orphaned processes that do not listen on a port, listed with command, project, memory, and which agent started them.

1
2
417
green tests say the suite is green. they do not say the change is in scope, secret-free, reversible, or that it ran for real. six checks sit between green and done. a receipt you can verify without uploading source. nyk.dev/products/release-rec…
Screenshot from the Release Receipt page, "The six checks between green and done." Six cells: behavior-proof, scope-fidelity, supply-chain, secret-safety, rollback-proof, run-evidence, each with a one-line definition. Caption: each check is a plain-text contract you own; the CLI runs a deterministic first pass; all six are the Six-Part Release Receipt.

ALT Screenshot from the Release Receipt page, "The six checks between green and done." Six cells: behavior-proof, scope-fidelity, supply-chain, secret-safety, rollback-proof, run-evidence, each with a one-line definition. Caption: each check is a plain-text contract you own; the CLI runs a deterministic first pass; all six are the Six-Part Release Receipt.

1
292
reviewers stop seeing the same stock phrases after a week of reading them. a catalog of 148 tells does not get tired: line, rule, excerpt, and exit 1 at 40 so CI can stop the draft. that's what copylint is for. a linter, not a detector.
Screenshot from the Copylint site, section "A linter, not a detector". Comparison table: Copylint vs AI detectors vs humanizers on the question it answers, what you get back, whether your words stay yours, and whether the same draft scores the same every run. Copylint column: findings with line and rule plus suggested fixes; change only where you accept a fix; every run, everywhere.

ALT Screenshot from the Copylint site, section "A linter, not a detector". Comparison table: Copylint vs AI detectors vs humanizers on the question it answers, what you get back, whether your words stay yours, and whether the same draft scores the same every run. Copylint column: findings with line and rule plus suggested fixes; change only where you accept a fix; every run, everywhere.

240
a decision you can't check later becomes a story you tell afterwards. before you act on a council verdict, write down the prediction, an owner, a review date and the evidence that would change it. then mark the result. don't rewrite the reasoning. github.com/0xNyk/council-of-…
Card titled "before you act on the verdict, write down", from the council-of-high-intelligence README section "Track the outcome". 01 the prediction, what you expect to happen. 02 the owner, who checks it. 03 the review date, when it gets checked. 04 the evidence, what would change the recommendation. Then: at the review, mark it confirmed, revised, reversed or inconclusive. Don't rewrite the original rationale.

ALT Card titled "before you act on the verdict, write down", from the council-of-high-intelligence README section "Track the outcome". 01 the prediction, what you expect to happen. 02 the owner, who checks it. 03 the review date, when it gets checked. 04 the evidence, what would change the recommendation. Then: at the review, mark it confirmed, revised, reversed or inconclusive. Don't rewrite the original rationale.

1
1
3
284
if your agent says done and there's nothing you can rerun to check it, you have a claim. the agent os kit is how i close that: a task contract, evidence a person can reproduce, a failure record that changes the next run. nyk.dev/products/agent-os-ki…
Screenshot from the Agent OS Kit page, "Failure patterns: Fix the operating loop, not the next prompt." Three cards: Rules drift, where project commands, boundaries and decisions live in prompts, chat history and operator memory. Completion is a claim, where the agent says the work is done but no reproducible receipt proves the real user path or release checks. Retries learn nothing, where a failed run gets more context and another attempt without changing the conditions that caused the failure.

ALT Screenshot from the Agent OS Kit page, "Failure patterns: Fix the operating loop, not the next prompt." Three cards: Rules drift, where project commands, boundaries and decisions live in prompts, chat history and operator memory. Completion is a claim, where the agent says the work is done but no reproducible receipt proves the real user path or release checks. Retries learn nothing, where a failed run gets more context and another attempt without changing the conditions that caused the failure.

1
1
341
Load more