umararshad4/honey-stack / skills/honey
honey
UnverifiedPick the honey-stack skill or flow for a situation, or switch response modes: direct, explain, balance, status, off, on, default.
honey
Router and mode switch for honey-stack.
- Mode words (
direct,explain,balance,status,default direct, …): apply them per the Modes section and reply with one status line. update hooks: followreferences/update-hooks.mdfor a one-time Honey release check and Codex hook review.- A situation ("I have a bug report and a half-written plan"): name the skill or flow that fits, say why in one line, then run it.
- Nothing: show the main flow and the situation table below, trimmed to what fits the current repo.
Main flow: idea to shipped
grill ──> prototype? ──> plan ──> execute ──> verify ──> review ──> ship
grill: pin down intent and the product calls the repo can't answer. Quick or small work skips straight to building.prototype, only when a question needs a runnable answer (behavior, timing, a layout you have to see). Bridge out and back withhandoffif it needs its own session.plan: a session plan for work that fits one or two sessions, or a spec plus tracer-bullet tickets for a multi-session build.execute: drain the plan, tickets, orTODO.mdwithout check-ins. For tickets, one fresh session per ticket.verifyafter each item,reviewon the finished diff,shipwhen the user asks to commit, push, or open a PR.
Bugs merge onto the flow at debug, which hands its fix to verify.
Situation to skill
| Situation | Skill |
|---|---|
| Vague idea, missing product decision, "grill me", "brainstorm" | grill |
| Question only running code can settle; throwaway sketch | prototype |
| Multi-file or multi-session work; plan, spec, PRD, tickets | plan |
| "continue", "next", "do the TODOs", "run the plan" | execute |
| Broken, throwing, failing, flaky, slow; a fix that didn't work | debug |
| Writing a feature or fix test-first at an agreed seam | tdd |
| About to say done, fixed, or passing | verify |
| Diff or PR ready for a second look; self-review | review |
| Commit, push, PR, release, changelog | ship |
| Deploy to a cloud/PaaS, write Docker/K8s/IaC, fix a failing deploy or pipeline | deploy |
| Design the test matrix and hard edge cases for a feature before writing tests | test-design |
| Set up a monorepo, micro-frontend, microservices, pre-commit hooks, lint/format/CI | tooling |
| Schema design, migrations, indexing, slow queries, N+1, transactions | database |
| Profiling, hot paths, caching, load testing, backend perf with before/after numbers | perf |
| Logging, metrics, tracing, error tracking, SLOs, alerting, instrument a service | observability |
| Design or version a REST/GraphQL/gRPC API, error contracts, pagination, idempotency | api-design |
| Safely upgrade dependencies, patch a CVE, handle a breaking change or EOL | upgrade |
| A production outage: triage, mitigate, roll back, communicate, postmortem | incident |
| A skill misfired or a lesson should stick; improve honey-stack itself | rsi |
| UI, layout, styling, responsive, design-system work | frontend |
| Mocked product UI illustrations or animated cursor demos | stageshow |
| Application animation, motion audit or dropped frames | frontend (motion references); review or perf for their respective checks |
| Drive the real app to find bugs before users do | qa |
| Auth, sessions, secrets, payments, user data, webhooks | security |
| Behavior, setup, commands, or APIs changed; docs drift | docs |
| Text reads like AI wrote it; tighten a message, post, prompt or doc | unslop |
| Thread is long or branching; carry state to a fresh session | handoff |
| Codebase hard to change or test; deepen modules, find seams | architecture |
Name the domain terms, keep CONTEXT.md and ADRs | domain |
| Learn a concept over several sessions | teach |
| Write or edit a skill | write-skill |
| A design call needs a named principle (root cause, idempotency, boundaries) | principles |
| Need a library's real source or docs for a declared dependency | opensrc |
| Research a technical/product question against primary sources | research |
| Create or maintain this project's launch/drive/evidence harness | project-verify |
| Qualify issues, deduplicate reports or prepare an actionable agent brief | triage |
| Model a queue, retry or workflow invariant and its counterexamples | statechart |
| Show honey's measured benchmark gains | gain |
| Understand a shipped app, binary, APK or site you don't have source for (authorized) | reverse-engineer |
When two rows fit, take the earlier one in the main flow: settle intent before planning, plan before executing.
Overlaps — which one wins
debugvsincident: local/non-urgent root-cause →debug; production is down or degraded right now →incident(mitigate first, thendebugfor the root cause).perfvsdatabase: a slow query, N+1, index, or migration →database; profiling, caching, load testing, or a non-DB hot path →perf. A slow endpoint caused by SQL isdatabase; by CPU/serialization/IO isperf.tddvstest-designvsqa: what cases to test →test-design; writing them test-first →tdd; driving the real browser/UI →qa.reviewvsverify: judging someone's diff/PR →review; proving your own change is actually done before you claim it →verify.shipvsdeploy: commit/push/PR →ship; getting it running on a server/cloud →deploy.securityvsupgrade: a design/code security concern or audit →security; patching a vulnerable dependency →upgrade.docsvsunslop: writing/updating project docs →docs; tightening any prose of AI tells →unslop.project-verifyvsverifyvsqa: author/maintain a project harness →project-verify; prove a particular change →verify; browser interaction →qa. Reuse the project's harness when it exists.triagevsdebug: qualify a report and prepare a brief →triage; investigate a confirmed failure →debug. A proposed fix is not verification.researchvsopensrc: broad primary-source investigation →research; installed dependency API context →opensrc.frontendvsstageshow: application motion and UI quality →frontend; mocked product illustrations and cursor demos →stageshow.qaowns stress/mobile evidence.
Modes
One response style at a time. All modes keep the same task sizing, verification, and safety rules.
| Word | Effect |
|---|---|
direct | Straight to the point, concise responses. |
explain | Comprehensive explanations with useful context, reasoning, and examples. |
balance | Moderate detail: answer plus key reasoning, neither comprehensive nor too brief. |
status | Print the active response mode and whether Honey is paused. |
off / on | Pause the whole layer (bootstrap, hooks, routing) / resume it. |
default <mode> | Save the mode for new sessions, e.g. honey default direct. |
The default is direct. Session switches last until changed or the session ends. The indicator includes the active choice, for example 🍯 HONEY · balance. Explicit requests for more or less detail take priority over the selected style.
The hook does the switching. Send honey explain or /honey balance (Codex: $honey balance) as the whole message and the prompt hook applies it before the model runs. If this skill receives mode words instead, the hook didn't fire: honor the mode in conversation for the rest of the session and say in one line that nothing was saved.
Existing saved terse settings read as direct, and normal as balance. Old scope/process choices no longer change the workflow. Saving a new default writes mode in the global config and removes the old modes field.
Explain active rules
When asked what Honey applied, identify the relevant bootstrap, skill or learned preference and the evidence of its use. honey doctor --agent codex --json (or --agent claude) reports observed hook delivery; honey task status --json reports task evidence. Instruction loading alone does not prove enforcement. A response-mode change only changes presentation; preserve learned preferences unless the user explicitly asks to change or forget them.
Context hygiene
- One unbroken thread for grill → plan, so the plan argues from the same thinking. Compact or clear after the plan is written, not during grilling.
/clear(Codex:/new) between unrelated tasks. Leftover context from task A steers task B.handoffforks;/compactcontinues. Usehandoffwhen you want a fresh session but need this one's state (a prototype detour, a thread nearing its smart zone, the next ticket). Use/compactat a phase boundary inside one task when losing verbatim history is fine; never mid-debug or mid-grill, where the details are the work.- Fresh session per ticket. Start each with
execute, the spec, and that one ticket. - Reasoning degrades well before the window is full. When a long thread starts repeating itself or losing decisions, write a
handoffand continue fresh rather than pushing on.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 319k | Updated | Claude CodeCodex | Verified | Install |