Claudboard

Claudboard — Project Onboarding Agent

Analyzes a repository and generates Claude Code artifacts that let you work on it immediately: CLAUDE.md, rules, and full-scope skills tailored to the project's actual patterns.

Where to run /analyse

single repo  → run inside the repo
monorepo     → run at the repo root
workspace    → run at the workspace directory
               (the parent folder that holds ONLY the related repos
               — not a generic ~/Projects or ~/code folder)

If you run from the wrong level, re-run from the right one. The catalog regenerates in seconds.

Hinting tracker/repo backend for /claudboard-workflow

When both a Jira MCP and a Bosch T&R MCP (or both ADO and GitHub MCPs) are configured, /claudboard-workflow will normally prompt once to ask which backend to target. Include a hint in your invocation text to skip that prompt:

Intent	Include in invocation	Example
Target Atlassian Jira	jira or atlassian	/claudboard-workflow use jira
Target Bosch T&R	tr, t&r, track, or bosch	/claudboard-workflow for bosch tr
Target Azure DevOps	ado, azure, or devops	/claudboard-workflow ado project
Target GitHub	github or gh	/claudboard-workflow github repo

Hints are matched as whole words (case-insensitive). An invocation with no hint produces at most one prompt; an invocation with a valid hint produces zero.

Phase 0: First-run permission setup

Before dispatching to any sub-skill, check whether the recommended permissions are already installed:

1. Read .claude/settings.json (if it exists).
2. Look for the key _claudboard_permissions_version in the JSON:
   • Present (any value): Skip the prompt — permissions were already handled. Proceed to sub-skill dispatch.
   • Absent: Show the one-shot offer below, then proceed.

One-shot offer (shown only when marker is absent):

Add claudboard's recommended permissions to .claude/settings.json? This eliminates ~20 prompts per analysis run. [y/n]

YES branch:

1. Read references/recommended-permissions.json to get the permissions.allow array.
2. Read .claude/settings.json (or start with {} if missing).
3. Merge: take the existing permissions.allow array (if any) and union it with the bundle's array — deduplicate, preserve all existing entries, never delete anything.
4. Write back .claude/settings.json with the merged permissions.allow and the marker "_claudboard_permissions_version": "1".
5. Print: ✓ Recommended permissions added to .claude/settings.json
6. MUST NOT modify any key other than permissions.allow and _claudboard_permissions_version.

NO branch:

1. Write "_claudboard_permissions_version": "1-declined" into .claude/settings.json (stamp only — do not touch any other key).
2. Print the bundle contents as a copy-paste fallback:

   To add these manually later, copy the permissions.allow entries from skills/claudboard/references/recommended-permissions.json into .claude/settings.json. To re-trigger this offer, delete the _claudboard_permissions_version key and re-run any claudboard command.

Workflow

Onboarding is a two-phase process. Run each phase separately for best results.

Phase 1: Analyse

/analyse [path]

Scans the project (read-only), detects patterns, quality signals, and anti-patterns, then presents an analysis report with proposed artifacts. Saves the report to .claude/reports/claudboard-analysis.md.

At the end of analysis, you'll be asked whether to continue with generation or do it later.

Phase 2: Generate

/generate

Reads the saved analysis report and generates production-ready .claude/ artifacts: CLAUDE.md, rules (with paths: frontmatter), and full-scope skills (SKILL.md + references/ + scripts/).

Recommended: Run /generate in a fresh Claude Code session. The analysis phase fills context with discovery data that isn't needed during generation — a clean session gives better output.

Additional commands

• /refresh — Delta updates for projects that already have .claude/ artifacts. Identifies what's new, stale, or missing and updates only what changed.
• /techdebt — Deep tech debt analysis. Produces module-grouped, ticket-ready reports with severity, effort, and fix suggestions.
• /claudboard-workflow — Generates a tailored .claude/skills/feature-workflow/ skill into the target project from the analysis report. Requires /generate to have run first. Also triggered by: "set up feature workflow", "install start-feature skill", "generate feature-workflow", "configure feature workflow for this repo".
• /claudboard-workspace-init — Bootstrap a workspace meta-repo: creates a sibling git repo, migrates existing .claude/ contents, and symlinks the workspace root's .claude/ to the meta-repo. Run once per workspace from the workspace root. Also triggered by: "bootstrap workspace", "set up workspace meta-repo", "share my .claude across the team".
• /claudboard-workspace-link <url> — Teammate bootstrap: clones the workspace meta-repo and creates the symlink. Run after the first developer has run /claudboard-workspace-init. Also triggered by: "join workspace", "link to workspace meta-repo", "clone workspace meta-repo".

Reference Files

Sub-skills load these shared references as needed:

File	Used by	Purpose
references/stack-detectors.md	analyse, techdebt, refresh	Shared detection heuristics, infra signals, monorepo detection
references/workflow-signals.md	analyse	Dispatcher for cross-service edge schema + sub-catalog pointer table
references/edges/sync-rpc.md	analyse	REST, gRPC, tRPC detection (Java + TS)
references/edges/messaging.md	analyse	Kafka, AMQP, JMS, SNS/SQS, Solace, Azure Service Bus detection
references/edges/streaming.md	analyse	WebSocket, SSE, RSocket detection
references/edges/graphql.md	analyse	GraphQL client + server detection
references/patterns/architectural.md	analyse	Saga, CQRS, outbox, BFF, circuit breaker, schema registry detection
references/stack-detectors-java.md	analyse, techdebt	Java/Kotlin Wide Scan patterns (7 categories)
references/stack-detectors-typescript.md	analyse, techdebt	TypeScript/JavaScript Wide Scan patterns (7 categories)
references/stack-detectors-python.md	analyse, techdebt	Python Wide Scan patterns (7 categories)
references/stack-detectors-go.md	analyse, techdebt	Go Wide Scan patterns (7 categories)
references/stack-detectors-rust.md	analyse, techdebt	Rust Wide Scan patterns (7 categories)
references/stack-detectors-dotnet.md	analyse, techdebt	.NET/C# Wide Scan patterns (7 categories)
references/pattern-catalog.md	analyse, techdebt	Architecture patterns + anti-patterns catalog
references/quality-signals.md	analyse, refresh	Quality scoring, rule depth, skill triggers
references/claude-md-template.md	generate	CLAUDE.md generation template
references/rule-templates.md	generate	Rule file templates per language
references/skill-generation.md	generate	Full-scope skill generation guide

The techdebt skill also has its own references in ../claudboard-techdebt/references/: code-smell-catalog, design-debt-patterns, perf-debt-patterns, arch-debt-patterns, severity-matrix, and report-template.

Per-task cost reporting (opt-in)

See the "Per-task cost reporting" section in README.md for the opt-in Stop hook that emits a cost line after every /analyse, /generate, /refresh, or /techdebt run. The hook uses scripts/compute-cost.sh and scripts/stop-hook.sh and costs $0 in API tokens.