Every tool, every skill.

The toolchain.

Six command-line tools installed by the bootstrap script — Claude Code (the agent itself), GitHub, Google Workspace, Cloudflare, Salesforce, and Databricks. Each one unlocks a domain: agent runtime, source code, productivity, infrastructure, sales, data.

Claude Codeclaude— the agent runtime itself

The agent runtime. Everything in this OS runs through claude. It boots into the repo, reads the identity files, loads the MCP servers, registers the hooks, and exposes the skills as slash commands. Without this, you have a folder of markdown.

Key surfaces: claude (start a session), claude --resume (pick up the last conversation), claude mcp list (see what's connected), /plugin (manage plugin marketplaces).

GitHubgh— official GitHub CLI

Manage repos, PRs, issues, gists from the terminal — and from Claude. The agent uses this for everything from gh repo create in /new-project's code-repo branch to opening PRs and reading issue threads. Scopes granted: gist read:org repo workflow.

Most-used from inside Claude: gh pr create, gh pr view, gh repo clone, gh issue list.

Google Workspacegws— Gmail, Calendar, Drive, Docs, and the rest

The biggest tool in the kit. gws talks to every Google Workspace surface — Gmail, Calendar, Drive, Docs, Sheets, Tasks, Slides, Forms, Meet, Chat — and backs ~40 of the 84 skills. Granted scopes: calendar, drive, gmail.modify, documents, spreadsheets, presentations, tasks (plus auxiliary). Chat and Forms require re-auth to add.

If gws auth status reports authenticated, every /gws-* and /recipe-* skill works. This is the workhorse.

Cloudflarewrangler— Workers, Pages, KV, D1, R2

Deploy and manage Cloudflare Workers, Pages, KV, D1, R2. Used by the /samba-publish skill to deploy HTML artifacts (briefings, dashboards, prototypes) to internal-only URLs with SSO gating. Also the path for any custom edge logic.

Most-used: wrangler pages deploy, wrangler deploy, wrangler tail.

Salesforcesf— orgs, SOQL queries, metadata deploys

Query Salesforce orgs, deploy metadata, run SOQL. Most useful if you're working on CRM integrations or pulling deal pipelines into briefings. If sf org list shows decryption errors, re-auth: sf org login web -a <alias>.

Most-used: sf data query --query "SELECT ...", sf project deploy start, sf org open.

Databricksdatabricks— workspace, jobs, SQL queries

Workspace management, job runs, SQL queries against Databricks. Useful for data work — pulling tables, kicking off ETL jobs, fetching dashboards. Pass -p <profile-name> or set DATABRICKS_CONFIG_PROFILE if the default profile is invalid (common on shared workstations).

Most-used: databricks workspace ls, databricks jobs run-now, databricks sql query.

Model Context Protocol servers.

Five integrations registered in .mcp.json at the repo root — Gemini Vision, Exa Search, Slack, Atlassian (Jira + Confluence), and Figma. One runs locally; the other four authenticate through your browser on first use. They extend Claude with domain-specific tools: vision, web search, messaging, ticketing, design.

Gemini Visiongemini-vision— local stdio · vision + multimodal

Local Node MCP server backed by Google's Gemini API. Seven tools: image analysis, OCR (text extraction from screenshots), multi-image batch, image comparison, smart filename suggestion, document analysis (PDF / DOCX / TXT), and video analysis (local files or YouTube URLs). Uses the free Gemini tier (15 requests per minute).

Backs /find when it runs OCR over screenshots in your Inbox. Backs the meeting-prep flow when you point it at a YouTube recording. Auth: requires GEMINI_API_KEY in your shell — get one free at aistudio.google.com/apikey.

Exa Searchexa— HTTP · web search

Web search integration from Exa. Discipline: only web_search_advanced_exa is used — web_search_exa and web_fetch_exa are deprecated in this workspace. Categories supported: company (homepages + metadata), news, people (public LinkedIn), personal site, no-category (general).

Backs the /company-research and /people-research skills. Token-isolation rule: never run Exa searches in the main conversation context — always spawn a Task agent that runs Exa and returns distilled output, otherwise web noise pollutes your working context. Auth: handled by Exa's HTTP transport on first call.

Slackslack— HTTP · messaging

Read, send, and search Slack messages from inside Claude. Capabilities: read channels, read threads, send messages, search public + private (with permission), schedule messages, read user profiles, manage canvases (read / create / update). Verified working via slack_search_channels returning live results.

Known limit: no channel CRUD — you cannot create, archive, rename, set-topic, invite-user, or pin-message via MCP. Channel CRUD stays in the Slack UI. The MCP handles the messaging surface only. Auth: dynamic-client-registration OAuth on first use (clientId + callbackPort defined in .mcp.json).

Atlassian (Jira + Confluence)atlassian— HTTP · tickets + wiki

Talk to Jira and Confluence from Claude. Capabilities: search/create/edit Jira issues, transition issues through workflow states, add comments + worklogs, create issue links, read/create/update Confluence pages, fetch page descendants, search via CQL, manage inline + footer comments. Powers /scaffold-engineering-project, /confluence-publish-markdown, and /jira-decompose-epic.

Auth: browser OAuth on first call. After OAuth, getAccessibleAtlassianResources lists the sites you can reach.

Figmafigma— HTTP · design

Read Figma designs and (with the right scopes) write back. Capabilities: get design context for a node (returns code + screenshot + hints), get metadata, get screenshots at any zoom level, work with FigJam boards (read + create diagrams), manage Code Connect mappings between Figma components and your codebase components, upload assets, generate new Figma files.

Auth: browser OAuth on first call. Verify with figma__whoami.

Event-driven hooks.

Six hooks wired through .claude/settings.json. Each fires on a specific Claude Code event. This is the recap — the Architecture page covers Skills and Connections in depth; hooks live here as a pure reference list.

intent-detector— UserPromptSubmit

Reads your prompt before Claude does. Maps natural-language intent ("brief me", "what's stale?") to skills (/briefing, /prune-projects) so you don't have to memorize slash commands.

status— StatusLine

The persistent footer in Claude Code — project, branch, token count, model, working directory. Always visible.

tldr-read— PreToolUse (Read)

File-reading optimization. Short-circuits large file reads to a pre-computed TLDR summary to save tokens. Especially useful for big design docs and CSVs you reference often.

post-edit-diagnostics— PostToolUse (Edit)

Runs lint / typecheck after Claude edits a file. Surfaces errors back so Claude can fix them in the same turn instead of waiting for CI.

pre-compact— PreCompact

Snapshots context before Claude Code's auto-compaction kicks in. Preserves decisions, project state, and in-flight commitments that a generic summary might lose.

auto-handoff-stop— Stop

Generates a handoff doc into memory/YYYY-MM-DD.md when you end the session. Tomorrow's session boots warm instead of cold-start.

Each hook is one short script doing one job. Wired through .claude/settings.json with $CLAUDE_PROJECT_DIR so they travel with the repo. Lib helpers in .claude/hooks/lib/ handle shared concerns like resolving project root and writing handoff docs.

84 skills, grouped.

Slash commands you can invoke from any prompt. Too many to list individually here — organized by category below. Each skill lives in .claude/skills/<name>/SKILL.md and is auto-loaded at session start.

Project lifecycle — 5 skills

Scaffolding, archiving, triage. The skills that move work through the PARA folder model.

Chief of staff — 6 skills

Daily orientation, recall, person-tracking. The skills you'll run every morning and every meeting.

Autonomous + research — 6 skills

The CCv4 pipeline. Long-running work — autonomous SDLC loops, deep research, pre-implementation failure analysis, code + security review.

Handoff + bootstrap — 5 skills

Session continuity and fresh-fork onboarding. /bootstrap walks new users through identity setup; /bootup warms a project for daily use.

Design — 26 skills

HTML / visual artifact generators. Decks, dashboards, landing pages, mobile UI, posters, marketing emails, video, audio. All read the active DESIGN.md at repo root — swap brand with /use-design <brand>.

Persona workflows — 10 skills

Role-modeled workflows. Each loads a persona and composes the right primitive skills (calendar, email, docs, projects) for that role's daily flow.

Google Workspace — ~40 skills (gws-* + recipe-*)

Atomic gws-* primitives plus composed recipe-* workflows. Covers Gmail, Calendar, Drive, Docs, Sheets, Tasks, Slides, Forms, Meet, Chat. The longest tail of the skill library — only a sample below.

Publishing + engineering — 4 skills

Site building and ticket scaffolding. Wraps the Atlassian + Cloudflare MCPs in higher-level workflows.

Meta + harness — 9 skills

Skills for managing the harness itself. Author new skills, sync indexes, tune permissions, configure keybindings, set up recurring agents.