Wishes

Point kazam at a folder with your real context, let your agent read it, get back a populated YAML. No terminal interviews, no file parsing in kazam.

The flow

Scaffold, drop, grant

A wish is a workspace template plus the prompt that tells your agent how to turn that workspace into a valid kazam YAML. kazam scaffolds the workspace and shells out to your agent — it doesn't parse PDFs, chunk large files, or do anything the agent is already good at.

1. Make the wish

kazam wish deck with no wish-deck/ folder present scaffolds one: questions.md for structured prompts, README.md for what to drop in, and reference/ with the schema + a worked example (version-matched to the binary).

2. Add your context

Fill in any parts of questions.md you already know the answer to. Drop your real material into the folder — meeting notes, transcripts, last quarter's deck, PDFs, metrics dumps. Anything the agent can read is fair game.

3. Grant it

kazam wish deck a second time. kazam picks the first agent on your $PATH (Claude, Gemini, Codex, OpenCode), runs it from inside the workspace, and writes deck.yaml. The agent reads the workspace with its own file tools.

First wish

kazam wish deck

A 7-slide QBR / strategy-review deck. The agent writes every slide — cover, context, wins, challenges, learnings, next, the ask — from whatever you drop into the workspace. No placeholder copy.

kazam wish deck                          # 1st run: scaffold wish-deck/
# …fill in wish-deck/questions.md, drop your context files…
kazam wish deck                          # 2nd run: grant it
kazam wish deck --yolo                   # skip the workspace, agent invents everything
kazam wish deck --yolo "about me"        # YOLO with a topic (agent pulls from context)
kazam wish deck --agent claude           # force a specific agent
kazam wish deck --dry-run                # print the grant prompt
kazam wish deck --stdout                 # portable wish spec (markdown)
kazam wish deck --out quarterly.yaml     # pick the output path
kazam wish list                          # see every wish
🎲 YOLO mode

Skip the workspace entirely with kazam wish deck --yolo. The agent invents every slide — no questions, no context files, just vibes. Pass an optional topic (--yolo "about me", --yolo "history of coffee") or leave it bare for a surprise. Great for demos, warm-ups, or when your agent already knows you well enough to write the deck.

Applications

First-class in every agent

kazam auto-detects the first agent on your $PATH and shells out using its native non-interactive mode. No skill installer, no per-agent plugin format. Pass --agent <name> to force a specific one.

Claude Code

`claude -p` — best for agents that want to use Read tools on your context.

Gemini CLI

`gemini -p` — Google's CLI, same one-shot prompt shape.

Codex

`codex exec` — OpenAI's Codex CLI, subcommand form.

OpenCode

`opencode run` — SST's open-source agent CLI.

No CLI agent? No problem.

universal

Scaffold the workspace with `kazam wish deck`, fill in questions.md, then `kazam wish deck --dry-run` prints the full grant prompt. Paste it into ChatGPT, Claude.ai, Gemini web, Cursor, Aider, or any other LLM context window — save what comes back as `deck.yaml`.

The catalog

Every wish

deck ships today. The rest land one at a time — each one a Friday drop for a minor release. More wishes get added as patterns emerge in the wild.

kazam wish deck

shipped

QBR / strategy-review deck (7 slides). Cover, context, wins, challenges, learnings, next, the ask.

kazam wish brief

planned

Incident brief / meeting agenda. Document shell, print-optimized.

kazam wish landing

planned

Indie-hacker landing page with hero, features, stats, CTA.

kazam wish api-ref

planned

OpenAPI-style endpoint documentation.

kazam wish dashboard

planned

Dashboard + `sync.sh` reference script. You own the fetcher; kazam ships zero connectors.

kazam wish post-mortem

planned

Structured incident post-mortem — timeline, impact, learnings.

kazam wish resume

planned

Personal CV / resume riff.

kazam wish roadmap

planned

Public product roadmap with timeline and status.

Next
Next up → Components

Browse the component catalog to see what the rendered output can look like. kazam agents prints the full authoring guide — it's also what kazam drops into every wish workspace as reference/kazam-schema.md.