AI dev tools, battle-tested in production. Skills built and used daily by a working team — shared for the rest of the Claude Code community.
This is a Claude Code plugin marketplace maintained by hyperdrive.bot. Every skill in here is something the team actually uses — no toy demos, no AI-slop. Showcase repo: no support promised, PRs welcome.
/plugin marketplace add hyperdrive-bot/marketplaceThen install whichever plugins you want:
/plugin install post-mortem@hyperdrive-bot
/plugin install reflect@hyperdrive-bot
/plugin install transcript@hyperdrive-bot
/plugin install watch-jam@hyperdrive-bot
/plugin install paseo-bundle@hyperdrive-bot
/plugin install branded-pdf@hyperdrive-bot
/plugin install diagram-design@hyperdrive-bot
/plugin install session-naming@hyperdrive-botGenerates branded, multi-page post-mortem PDFs from incident facts. The opinionated bit: smoking-gun rigor. Every claim in Fault/Detection/Response/Recovery/Root-cause must carry evidence (CloudWatch line + UTC timestamp, commit SHA, support ticket ID, AWS request ID, query result). The 14 banned hedging words (likely, probably, may have, appears to, ...) are caught by a linter before render — you can't ship a PDF that says "we believe the deploy may have caused it."
- Five-whys discipline: must end at a systemic cause, not a person
- Branded templates registry:
./templates/<your-brand>/proposal.html— copy and customize - Chrome-headless render loop with page-overflow detection (splits at natural boundaries; never edits CSS to shoehorn content)
- Ships with a neutral default template (charcoal + indigo); BYO brand to taste
Usage: /post-mortem --branding <name> --title "Auth Outage 2025-01-15"
A reflective learning loop. Analyzes accumulated MEMORY.md, checkpoint logs, and the last 5 session transcripts for recurring corrections and confirmations. Identifies promotable patterns (≥2 sessions, actionable, not already covered), drafts the exact file content for .claude/rules/<topic>.md or global CLAUDE.md additions, and returns numbered proposals you can approve all/some/none.
- Runs in a forked context — your main conversation stays clean
- Emotional redaction: strips profanity/frustration, keeps the technical signal
- Logs every promotion to
rule-promotions.mdso the next reflect doesn't re-propose - Resets the session counter after approved promotions
Usage: /reflect (or /reflect --dry-run)
Transcribes any audio file via OpenAI's Whisper API. Handles WhatsApp .opus voice notes by auto-converting to .mp3 first (Whisper doesn't accept opus directly). Optional summary mode for long recordings.
- Supports mp3, wav, m4a, ogg, opus, webm, flac, mp4
- Language hint via
--lang pt|en|es|...(ISO 639-1) - Auto-summary when transcription exceeds 500 words
- Offers to save as
<filename>.transcript.mdalongside the source
Requires: OPENAI_API_KEY env var, ffmpeg on PATH (for opus conversion).
Usage: /transcript path/to/voicenote.opus --lang pt --summary
Pastes a jam.dev/c/<uuid> URL → downloads the raw WebM via anonymous GraphQL → extracts evenly-spaced key frames with ffmpeg → Claude reads them with vision and tells you what the user actually did. Pairs beautifully with the Jam MCP server for full debug context (frames answer "what did the user see"; MCP answers "what did the browser do").
- Detail levels:
quick(3 frames),standard(6),thorough(12),many(24), or any integer - Frames spread 5%→95% of duration (avoids recorder chrome at start/end)
- Cached per-UUID under
$TMPDIR/watch-jam/— re-runs only re-extract frames - No auth required (anonymous share-URL access)
Requires: ffmpeg, node on PATH.
Usage: /watch-jam https://jam.dev/c/abc123... thorough
Six skills that lean on the Paseo daemon for managing background coding agents across providers (Claude Opus, GPT-5.4, etc.) and isolated git worktrees.
| Skill | When to use |
|---|---|
paseo |
Reference — load whenever you create agents or manage worktrees |
paseo-epic |
Heavy-ceremony orchestration: research → plan → adversarial review → phased implementation → audit → delivery. For all-night work. |
paseo-committee |
Stuck or looping? Two contrasting agents do parallel root-cause analysis, then you implement |
paseo-advisor |
Second opinion on the current task — single agent, no editing |
paseo-loop |
Worker/verifier cycle that runs until an exit condition is met (drive PR to green, drive tests to green, babysit a long process) |
paseo-handoff |
Transfer the current task to a fresh agent with full context (zero-context briefing) |
Requires: Paseo installed locally (the daemon manages agent lifecycle).
Usage: /paseo-epic --autopilot "implement the new auth flow end to end", /paseo-committee "why is this test flaky", etc.
Same render-and-overflow engine as post-mortem, but for any document (proposals, briefings, executive summaries, monthly reports, retrospectives). First-run wizard collects colors, fonts, and logo via AskUserQuestion, generates a brand template under ./templates/<name>/, and reuses it across all future documents.
- Brand registry:
./templates/<name>/proposal.html+render.sh— copy + edit CSS vars to add brands manually - Ships with a neutral default (charcoal + indigo) for zero-config first use
- Section components:
.metrics,.timeline,.phase-card,.highlight-box,.check-list, tables, badges - Chrome-headless render loop with page-overflow detection — splits at natural boundaries, never edits CSS to shoehorn
Usage: /branded-pdf --brand acme --title "Q1 2025 Proposal" — or --register-brand <name> to run the wizard standalone.
Generate technical and product diagrams as standalone HTML files with inline SVG. Types: architecture, flowchart, sequence, state machine, ER / data model, timeline, swimlane, quadrant, nested, tree, layer stack, venn, pyramid. Neutral editorial skin out of the box; first-run gate that prompts to customize the style guide from your brand site.
- Annotation-callout primitive for inline notes
- Optional sketchy variant (loose hand-drawn style)
- Complexity budget: target density 4/10 — earn every node, every connection
- Type-specific conventions live in
references/type-<name>.mdand load only when relevant
Usage: describe the system you want diagrammed; the skill picks the right type and emits a self-contained HTML file.
Five hooks that name your sessions automatically based on conversation topic, prefix done- when you say bye, and nudge /reflect every 10 sessions:
| Hook | Event | What it does |
|---|---|---|
session-id-inject |
SessionStart | Injects session ID + auto-rename directive into Claude's context |
session-counter |
SessionStart | Increments session counter; emits a reflect nudge when threshold hit |
session-auto-rename |
UserPromptSubmit | After 3 unnamed exchanges, forces Claude to rename NOW |
session-exit-rename |
UserPromptSubmit | Detects exit keywords (bye, tchau, done, exit, ...) and tells Claude to prefix done- |
session-done-on-end |
SessionEnd | Final prefix-done- pass on natural session end |
Plus a session-rename.sh utility that updates the JSONL custom-title record and the sessions-index so /resume picks up the new name.
Usage: install + use. The hooks run themselves. Pair with reflect for the full effect (reflect nudge → /reflect → durable rules).
This marketplace is curated. It does not include:
- Client-specific tooling — every skill here is generic and self-contained
- Hooks — those live in a future plugin once they're properly portable
- Rules — there are great ones in our day-to-day stack, but they're tightly coupled to our codebase. A separate
claude-code-good-practicesplugin may ship later with the generic ones (Lambda async patterns, CFN deletion-policy hygiene, etc.)
If you want to see the "real" super-repo with all the rules, agents, hooks, and the full BMAD-based workflow, it's on GitLab — but it's read-mostly and assumes a specific stack.
- v0.1 ✅ —
post-mortem,reflect,transcript,watch-jam,paseo-bundle - v0.2 ✅ —
branded-pdf(PDF engine wizard),diagram-design(13 diagram types),session-naming(hooks) - v0.3 — Rules plugin: generic Lambda / CFN / Git / npm rules from production incidents
- v0.3 — BMAD-extras plugin: Marra-customized personas not in
@hyperdrive.bot/bmad-sls-react(Tiririca, Elara, Spike, ...) - v1.0 — Polished landing page at hyperdrive.bot
PRs welcome but go in with low expectations:
- This is a showcase, not a community-maintained suite
- Bug reports and clean fixes — likely yes
- Big new features — probably better as your own marketplace
- Style/cosmetic changes — generally no
Skills, hooks, and rules added here must be:
- Generic — works in any codebase, not tied to one company's stack
- Self-contained — no internal dependencies, no private CLI requirements
- Battle-tested — used in real work, not "looked cool in a demo"
Built and maintained by the team at DevSquad. The opinions encoded in these skills (smoking-gun post-mortems, refactor-first orchestration, ban hedging in incident reports, no waitForTimeout in tests) come from real production scars. We figured shipping the tools was less work than re-deriving them.