caveman
claudeTalk like caveman. Cut filler. Keep technical details explicit.
22 skills · 1 command · 3 agents · hooks
Personal model-first routing for coding agents
Plugin dmnkstack in dmnkgrc/dotfiles. Commands read from the repo's manifest — 18 days ago.
architectSettle caller usage, types, interfaces, state ownership, and module boundaries before implementation. Use for architect this, design this, cross-cutting changes, concurrency, migrations, new APIs, or work where coding first could lock in the wrong shape.
arenaRun several independent attempts at the same non-trivial task, judge them against one rubric, and synthesize one result. Use for arena this, competing designs, difficult prompts, prototypes, or decisions where one attempt could lock in the wrong shape.
automate-meLearn durable working preferences from local Codex, Claude, Pi, and Cursor history and turn them into a personal Dmnkstack working style. Use when the user says automate me, asks to personalize Dmnkstack from past conversations, wants recurring correction patterns analyzed, or wants the router tuned
before-and-afterAdd existing screenshots or screen recordings to a GitHub pull request as a before/after or preview block. Use when a PR needs visual media attached to its description. Browser navigation and capture belong to agent-browser.
blast-radiusFind what a proposed or completed change could break outside its diff and prove the key safety assumption with running code. Use for blast radius, what could this break, risky small diffs, compatibility changes, schema changes, shared types, or hidden downstream consumers.
debugReproduce, diagnose, fix, and verify software failures. Use for bugs, regressions, crashes, flaky behavior, incorrect output, runtime anomalies, performance problems, failing tests, or requests to find a root cause. Keep the work read-only when the user asks only for diagnosis.
delegateRun work through another model or agent and collect its result into the parent session. Use when Dmnkstack selects a model unavailable to the current agent, the user asks for another agent or model, a panel needs independent judgments, work inside Conductor should happen in a new same-workspace sess
deliverPrepare and publish a completed software change through the user's requested git workflow. Use when the user explicitly asks to commit, push, open or update a pull request, babysit or watch a PR, address delivery checks or review findings, rebase a PR, says "full stop", or hands off a branch. This s
design-handoffTurn a Figma design, node URL, screenshot, or visual specification into project-native UI code and verify it in the running product. Use for Figma-to-code work, design implementation, visual parity, responsive states, component mapping, or UI changes where the design is the source of truth.
dmnkstackRoute software work across available project skills, task playbooks, models, and agent runners. Use when the user invokes Dmnkstack, asks for Pstack-style model routing, wants the best agent or model for a coding task, requests a cross-model panel, or continues a task already running in Dmnkstack mo
figure-it-outDesign and run a rigorous workflow when no narrower skill fits. Use for figure it out, large migrations, unusual multi-part tasks, ambiguous operational work, or long autonomous work that needs an auditable result when the user returns.
howTrace and explain how a subsystem, feature, request path, or module works. Use for code walkthroughs, ownership questions, layering questions, onboarding explanations, runtime flow, or architectural critique. Use why when the question asks for historical motivation.
interrogateRun an adversarial multi-model review of code, a diff, a plan, or a design. Use for interrogate, challenge this, stress test this, find blind spots, tear this apart, or review this change with independent models.
jam-triageTurn a Jam bug report, recording, screenshot, or shared reproduction link into a scoped reproduction and evidence-backed diagnosis. Use when a request contains a jam.dev link, mentions a Jam report, or supplies captured browser events, console logs, network requests, device details, or a user-record
observeInvestigate runtime behavior through installed observability and data skills or connectors. Use for Datadog logs, APM traces, metrics, Braintrust spans or evals, Temporal executions, Sentry events, BigQuery or Metabase evidence, production incidents, request IDs, workflow IDs, run IDs, and questions
recallRebuild the current state of prior work from local agent history, saved Dmnkstack state, git, and linked work records. Use when resuming a task, taking over a branch, asking what was decided, or recovering context after compaction or a long pause.
reflect-dmnkstackCapture a correction or task outcome as evidence for improving Dmnkstack. Use after a routed task, when the user says to remember a workflow preference, when a model route worked poorly or well, or when repeated friction suggests changing the personal working style.
setup-dmnkstackConfigure or repair Dmnkstack model routes, compatible agents, and Conductor or Herdr launchers. Use when installing Dmnkstack, changing model assignments, adding an agent runtime, validating model availability, or troubleshooting why a routed model cannot start.
show-me-your-workKeep a compact, reviewable decision and evidence trail during long or consequential work. Use when the user asks for an audit trail, steps away during an autonomous run, wants decisions recorded, or needs a reviewer to understand why each unit was kept or rejected.
swarmFan out independent slices of a task or race several approaches, then return one aggregate report. Use for swarm this, parallel coverage, package sweeps, independent research, test matrices, migrations with clean seams, or a requested model race.
tddFix a bug by writing the focused failing regression first. Use when the user asks for TDD, failing-before evidence, or a regression test, and when a bug has a cheap local executable test path. Skip when the only possible test would be broad, brittle, or mostly mocked.
teachTeach a codebase concept or subsystem until the user can reason about it independently. Use when the user asks to learn, understand deeply, be walked through a change, or connect runtime mechanics with historical rationale.
verifyProve that a software change works against the real artifact. Use when asked to verify, prove it works, test the result, check behavior, validate a fix, establish visual parity, or decide whether work is genuinely complete.
whyRecover why code, architecture, thresholds, flags, or product behavior has its current shape. Use for design rationale, code archaeology, regressions, incidents, rejected alternatives, historical tradeoffs, and questions whose answer requires source control or organizational evidence. Use how for me
Talk like caveman. Cut filler. Keep technical details explicit.
22 skills · 1 command · 3 agents · hooks
Persistent memory for Claude Code: the next session starts where the last one left off.
29 skills · 1 MCP server · hooks
tasteskill 2.5.1: professional company websites at the figma.com and x.ai level, with one section per screen, real photographs, real product UI, effects kept for tech and gaming pages, and a real-browser checker and finish gate.
1 skill
Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications
5 skills · hooks
Cross-session memory and token savings for coding agents.
6 skills · 1 agent · 1 MCP server · hooks
Package any website or local web build into a lightweight desktop app with Pake CLI.
1 skill
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.