[ironystock]

Summary

Final phase of the rememberize-cli bootstrap sprint. Adds the agent-ownership infrastructure (so a dedicated Claude Code agent can own this repo after merge) and upgrades the README from placeholder to a real install guide strangers can land on.

Six commits, each small:

1. CLAUDE.md — project context loaded into every Claude Code session (repo purpose, two-repo split, stack, commit conventions, pointers, dual-namespace MCP wiring).
2. AGENTS.md — owner-agent playbook (scope, cadence, out-of-scope, escalation via [MAIN-REPO] issues, session workflow with recall/remember, PR template, release workflow, starter backlog: F18 / binary signing / --vector flag).
3. README.md upgrade — tagline, install section (curl one-liner + raw fallback + Homebrew + Scoop + manual), quick start, skill pointer, license.
4. docs/agent-briefing.md — one-time first-run orientation walking the agent through AGENTS.md → CLAUDE.md → recall sprint context → recall friction → check issues → write first memory note → verify install → end-of-session note.
5. .mcp.json + .claude/settings.json — MCP wired to https://platform.rememberize.app/mcp with ${REMEMBERIZE_API_KEY} literal placeholder (documented: agent runtime supplies the real key, never committed); minimal Claude Code settings (Opus 4.7 model, empty hooks).
6. CONTRIBUTING.md — short contributor guide for strangers (scope, issue format, PR style, build/test, what lands vs needs prior discussion, server/protocol reports routed to [MAIN-REPO] issues).

Context

Phase A (PR #1 CLI source), Phase B (PR #2 bundle skill), Phase C (PR #3 release pipeline + install.sh) land alongside this. After all four merge, the repo is ready for first release.

This is the final phase

After merge + v0.1.0 tag, the repo is live and the dedicated owner agent can be spawned.

Brad's manual steps post-merge

1. Provision the agent's MCP connection in the rememberize dashboard. Client hint: rememberize-cli-agent. Dual grants on one API key:
   • {namespace: "rememberize-dev", memories: ["recall","search","list"], namespaces: ["list"]} — read-only, no remember/update/forget.
   • Full read/write on rememberize-cli-agent (the agent's own working memory).
2. Set REMEMBERIZE_API_KEY env var in whatever environment runs the agent (shell export, Claude Code harness config, or local .mcp.json override — whichever path the agent runtime actually picks up).
3. Tag v0.1.0 to cut the first release. goreleaser (Phase C) handles binaries + Homebrew/Scoop taps automatically.
4. Spawn the owner agent with docs/agent-briefing.md as initial input. It will walk itself through orientation and file starter issues.

Subtle decisions

• Module path / GitHub org: the sprint plan referenced ironystock/rememberize-cli but the actual repo lives at captured-ventures/rememberize-cli (per go.mod and the git remote). README + CONTRIBUTING use the actual path. AGENTS.md still refers to the private server repo as ironystock/rememberize per the plan — flag if you want that renamed too.
• .mcp.json env-var placeholder: kept the literal \${REMEMBERIZE_API_KEY} per the spec. Claude Code's MCP harness does not reliably interpolate env vars inline — CLAUDE.md documents three alternative ways for the runtime to inject the real key (shell export pass-through, local .mcp.json override, Claude Code settings injection). Never committed.
• .claude/settings.json minimalism: just {"model": "claude-opus-4-7", "hooks": {}}. No pre-tool hooks yet — the AGENTS playbook enforces make test lint by convention, and CI blocks merge. Easier to add hooks later than to debug a broken one on day one.
• .gitignore tweak: added .claude/settings.local.json and .claude/scheduled_tasks.lock so the shared settings.json commits cleanly but per-agent-instance state doesn't leak.
• No badges in README yet — per the plan, deferred to a later PR.

Anything we missed

One thing flagged for your judgment, not silently added: an issues/ directory or ISSUE_TEMPLATE. AGENTS.md mentions triaging issues/, and docs/agent-briefing.md tells the agent to check it, but there's no .github/ISSUE_TEMPLATE/ or issues/ inbox scaffolding yet. Given the stock gh issue list surfaces the GitHub Issues tab without any template, and templates are cheap to add in a follow-up PR once you've seen the first few reports, I left this out. Happy to add a basic bug-report + feature-request template in a quick follow-up if you'd prefer it pre-shipped.

Test plan

• CLAUDE.md under 150 lines (87 lines — well under).
• All referenced files exist (AGENTS.md, docs/agent-briefing.md, skills/rememberize-bundle/README.md from Phase B, CONTRIBUTING.md, LICENSE).
• .mcp.json parses as JSON; no real API keys.
• .claude/settings.json parses as JSON.
• .gitignore excludes only local/ephemeral agent state, not shared config.
• Post-merge: Brad provisions MCP key + tags v0.1.0 + spawns agent with docs/agent-briefing.md.

Risk

Low. Pure docs and config templates; no runtime code changes. Worst case, the agent runtime doesn't find the API key on first boot and the agent's first session has to fall back to non-MCP context — fixable by Brad pasting the key into a local .mcp.json override.

---

🤖 Generated with Claude Code