fordaaaa/panoply

---

Contents

• Install
• What you get
• What it won’t do
• Configuration
• MCP servers
• Agents
• Skills
• Contributing

Install

npx panoply init --global   # every project on the machine

Or npx panoply init inside one project to scope it there. Either way it detects Claude Code, opencode, or Cursor, installs there, never overwrites a command you wrote yourself, and configures nothing by default — everything starts in local mode.

In Claude Code you can install as an updatable plugin instead:

/plugin marketplace add fordaaaa/panoply
/plugin install panoply@panoply

Using something else? Open prompts/, copy a file, replace {{ARGUMENTS}}, paste it in.

What you get

> /cr-run standard

  Reviewing 84 files with 3 subagents (correctness · performance · security).
  Estimated ~180k input tokens. Continue? y

  🔴  src/auth/session.ts:142   Session token compared with ==, not timing-safe
                                 → use crypto.timingSafeEqual
  🟠  src/api/upload.ts:67      Unbounded read into memory; 2GB upload OOMs the worker
                                 → stream to disk, cap at configured limit
  🟡  src/db/pool.ts:23         Pool never drained on SIGTERM; deploys drop in-flight queries

  Verified 3 of 5 reported — dropped 2 I couldn't confirm at the cited line.
  Local mode: nothing filed. Want me to fix any of these, or start filing?

Review

Plan, verify, debug

Ship

Orient

The idea: the repo is the memory; context is disposable. Every command writes or reads a durable artifact, so nothing dies when context compacts or you switch tools. They compose through files (/map → /spec → /verify, /cr-run → /cr-fix) and still work standalone.

Make it yours: drop a markdown file in commands/ (or agents/, or skills/), run node build.mjs, and it exists in every agent. The bar in CONTRIBUTING.md: each addition must beat a plain prompt through structure — parallel subagents, a durable artifact, a verification loop, or a cheaper model.

What it won’t do

Worth knowing before you install something that can open pull requests.

• Won’t merge without earning it. autoclose defaults off; even on, five conditions must hold at once. Never passes --admin.
• Won’t treat your repo as instructions. Issue bodies and source text are data; fixes never touch CI config, workflows, lockfiles, or credentials.
• Won’t publish your vulnerabilities. Security findings stay on screen, with an offer of a private advisory.
• Won’t spend without asking. /cr-run deep prints a cost estimate and waits — and above 1500 files it refuses outright rather than taking a bare “yes”.
• Won’t assume. Not tests, not main, not a clean tree, not where gh points.

Findings carry a severity (🔴 Critical · 🟠 High · 🟡 Medium · 🟢 Low · ⚪ Trivial). Subagents self-score confidence 1–10 and report only 8+; every cited file:line is re-checked and unconfirmable findings dropped.

Configuration

Two questions, asked the first time you want something filed, saved to .panoply/config.md (filing: local | high-only | all, autoclose: off | on). Until then everything runs in local mode: report only, nothing filed, committed, or pushed. Say “reconfigure” or edit the file to change it.

MCP servers

Opt-ins pull in with npx panoply init --with . Everything comes from mcp/servers.json, rendered into .mcp.json, .cursor/mcp.json, and opencode.json. The GitHub server authenticates via GITHUB_MCP_TOKEN — see docs/mcp-auth.md (or skip it and just use gh).

Agents

opencode-only (Claude Code/Cursor have no equivalent): plan investigates read-only and writes a plan file; build executes it 1–2 files at a time with green tests and receipt lines. Sourced from agents/*.md, rendered into opencode.json’s agent block without touching agents you defined yourself. Starting template for your own repos: templates/AGENTS.md.

Skills

Claude Code only (opencode/Cursor have no equivalent) — sourced from skills/*/SKILL.md, rendered to .claude/skills/.

Running in CI / AWS

• CI: templates/github-workflows/panoply-review.yml runs /cr-run on every PR (never /cr-fix). Setup and cost notes in docs/ci-trigger.md.
• AWS: templates/docker-compose.floci.yml gives /verify and /cr-fix a local AWS emulator instead of skipping cloud paths as N/A. See docs/aws-testing.md.

Contributing

Edit commands/, skills/, agents/, and mcp/servers.json only; everything else is generated by node build.mjs. See CONTRIBUTING.md for the house style.

commands/                  ← the files you edit
├── _bootstrap.md          ← shared partials, pulled in via {{INCLUDE:_name.md}}
├── _preflight.md
├── _severity.md
├── _untrusted.md
└── cr-run.md, cr-fix.md         ← review
    map.md, spec.md, verify.md,  ← plan, verify, debug
    debug.md, prompt.md
    commit.md, test-gen.md,      ← ship
    ci-fix.md
    onboard.md, handoff.md       ← orient

skills/*/SKILL.md          ← Claude-Code-only, no opencode/Cursor equivalent
  (caveman, polyglot)

agents/*.md                ← opencode-only, rendered into opencode.json's agent block
  (plan, build)
      │
      │   node build.mjs
      ▼
.claude/commands/  .opencode/commands/  .cursor/commands/  prompts/
.claude/skills/
.mcp.json          opencode.json        .cursor/mcp.json

Credits

The parallel-subagent review structure and the severity/confidence scoring grew out of Anthropic’s MIT-licensed claude-code-security-review.