القائمة
Stack Honey onto Superpowers-style subagent workflows: dispatched subagents skip the session hook, so inject the levers into each dispatch prompt.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Compress-Cache-Retrieve huge repetitive array tool output before it hits context: keep a sample, cache the rest, leave a hash.
Compress-Cache-Retrieve for huge, repetitive array tool output (logs, scan results, time series, event streams) before it enters context. Keeps an informative sample — endpoints, anomalies/change-points, head/tail — drops the redundant rest to a local cache, and leaves a retrievable hash. Use when a tool returns a long uniform JSON array you must read but mostly skim, and the full set is one command away if needed. Lossy-but-recoverable.
Write less code and say less about it: YAGNI, stdlib-first, terse prose. Cuts agent token cost on every coding and writing task.
Write less code and say less about it. Applies YAGNI and stdlib/native-first so the agent writes the minimum code that needs to exist, and responds tersely — stripping filler, hedging, and pleasantries while keeping code, identifiers, and technical terms exact. Use whenever writing, modifying, refactoring, reviewing, or explaining code, or any response where output volume drives token cost — even if the user never says "minimal" or "concise". Especially in agentic coding, where the volume of generated code and prose runs up the bill.
Honey for plain Claude — the terse-prose core with no agent-harness features. Paste into a claude.ai Project's custom instructions, a Style, or an API system prompt. Strips filler, hedging, and pleasantries from every response while keeping facts, names, numbers, steps, and code exact. No tools required.
Read huge dense read-only text as PNG pages via pxpipe export: ~60-75% fewer input tokens. Lossy on exact strings; never for files you will edit.
| name | honey-superpowers |
| description | Stack Honey onto Superpowers-style subagent workflows: dispatched subagents skip the session hook, so inject the levers into each dispatch prompt. |
| homepage | https://github.com/Green-PT/honey-for-devs |
| license | MIT |
Honey is turned on by the session's SessionStart hook — which lives in your (the orchestrator's) context. Superpowers dispatches fresh subagents with isolated context; they never inherit the hook. So every implementer, reviewer, and fixer runs Honey-off: full-fat code (Lever 1 dark) and verbose reports (Lever 2 dark) — multiplied per dispatch. That's the leak. Close it by pasting the directive into the dispatch prompt itself.
The two skills already pull cost the same way Honey does — cheap model per task, file handoffs over pasted text, one-line narration. Don't fight them; stack on top. Honey adds what they don't: less code in the worker, terser reports both ways.
Any time you are about to dispatch a subagent under a Superpowers-style workflow:
subagent-driven-development — implementer / task-reviewer / fixer per taskdispatching-parallel-agents — one agent per independent domainexecuting-plans — parallel-session executionOn Claude Code with the Honey plugin, skip the paste: the plugin's
SubagentStart hook (hooks/honey-subagent.js)
injects the matching directive into every spawned subagent automatically
(reviewer-shaped agent types get the reviewer variant). The blocks below are for
other harnesses, or when the hook isn't installed.
Paste the matching block into the dispatch prompt body — inline text, not a Skill call (invoking burns a round-trip; the subagent only needs the rules).
Worker / implementer / fixer / parallel-domain agent:
Apply Honey: write the minimum code that needs to exist — YAGNI, stdlib/native
before custom; no speculative params, branches, or single-caller abstractions.
Never cut validation, error handling, auth, or anything the task asked for.
Keep code, identifiers, paths, and the brief's exact spec values verbatim.
Report terse: status, commits, one-line test summary, concerns. No narration.
Reviewer:
Report findings tersely: id · severity · file:line · one-line fix. Don't narrate
or restate the diff. Honey governs your prose only — never your verdict or
severity. Flag everything you normally would; raise false positives for the
controller to adjudicate. Do not suppress or downgrade a finding to save words.
Lever 1 (less code) and Lever 2 (less prose) compress what a worker emits. A reviewer emits judgments — compress its prose, never its verdict. This matches Superpowers' hard rule ("never tell a reviewer what not to flag") and Honey's own safety carve-out (keep exact anything asked for; auth/money/migrations/deletes stay explicit). A terser review is fine; a leniter review is a regression.
Never elide, in any variant: input validation at trust boundaries, error handling that prevents data loss, auth/escaping/secrets, accessibility basics, and anything the task or plan explicitly mandated. Spec values (numbers, magic strings, signatures, test cases) stay verbatim — that's Lever 2's "keep exact," and it's also what the brief is the single source of truth for.
The skill prompts Superpowers loads into context are input tokens you've already paid for — Honey can't out-compress them. They're lazy-loaded, so only triggered skills cost anything. The big realized savings are the cheap-model-per-task rule (Superpowers' own — don't skip it) plus Honey inside every dispatched subagent.