Generates a tailored `.claude/skills/feature-workflow/` skill into any project by reading the claudboard analysis report, gathering config from the user, and rendering Markdown templates with capability blocks and substitution variables. Produces a complete, project-aware feature-workflow skill (SKILL.md, agents, scripts, config.json) ready to drive `/start-feature` flows. Use when: /claudboard-workflow, "set up feature workflow", "install start-feature skill", "generate feature-workflow", "configure feature workflow for this repo", "add the feature workflow skill", "generate my start-feature skill", "create the feature workflow for me", "I want a start-feature skill", "bootstrap feature workflow", "generate workflow skill", "set up my workflow skill", "add start-feature to this project".

2026-06-083

claudboard-analyse

lukovicperisa/claudboard

Deep-analyzes a project repository: detects tech stack, architecture patterns, coding conventions, testing strategies, CI/CD pipelines, quality signals, and anti-patterns. Default mode produces a convention catalog (.claudboard/catalog.json) and a thin human-readable summary (.claude/reports/claudboard-analysis.md). Use --audit for per-service deep analysis (per-service reports, full Watch findings, cross-service Kafka graphs). Written to .claudboard/audits/ when --audit is passed. Use when: /analyse, "analyze this codebase", "scan this repo", "project health check", "audit this project", "what patterns does this repo use", "understand this project", "code quality review", "architecture audit".

2026-06-043

claudboard-generate

lukovicperisa/claudboard

Generates Claude Code artifacts (CLAUDE.md, rules, full-scope skills) from a claudboard convention catalog. Reads .claudboard/catalog.json as primary input; falls back to legacy .claude/reports/claudboard-analysis.md with one-time migration. Requires a prior `/analyse` run — does not analyse inline. Use when: /generate, "generate rules", "create CLAUDE.md", "bootstrap Claude for this repo", "generate skills for this project", "set up Claude Code for this project", "onboard this project".

2026-06-033

claudboard-refresh

lukovicperisa/claudboard

Refreshes Claude Code artifacts for a project that already has .claude/ context. Performs delta discovery against existing artifacts, identifies what's new, stale, or missing, and updates only what changed. Use when: /refresh, "update my rules", "refresh Claude context", "sync .claude/ with codebase changes", "what's changed since last scan", "are my rules up to date", "re-scan this project".

2026-06-033

claudboard-techdebt

lukovicperisa/claudboard

Deep tech debt analysis of a codebase. Produces a module-grouped, ticket-ready report with severity, effort estimates, and fix suggestions for every item. Detects code smells, missing design patterns, performance debt, and architecture flaws. Use when: /techdebt, "tech debt", "technical debt analysis", "what needs refactoring", "code quality report", "debt scan", "refactoring candidates", "find tech debt", "architecture debt", "performance issues in code", "generate refactoring tickets".

2026-05-203

feature-workflow

lukovicperisa/claudboard

Kick off intentional feature work with the full ticketed workflow: clarify scope, create a goal-oriented JIRA ticket, set up the branch, develop the solution, commit, and open a PR. Use this skill when the user says "start feature: X", "start a feature", "kick off work on X", or "full workflow for X". This is a deliberate mode the user enters — do NOT trigger for casual coding requests, experiments, spikes, or when the user just asks to "add X" or "implement X" without explicit feature framing. The phrase "start feature" is the signal.

2026-05-203

name

claudboard

description

Deep-analyzes an existing project or repository and bootstraps Claude Code's understanding by generating production-ready .claude/ artifacts: CLAUDE.md, rules with paths: frontmatter, full-scope skills (with references/ and scripts/), and memories. Goes beyond surface scanning to detect architecture patterns, coding conventions, testing strategies, CI/CD pipelines, infrastructure, good patterns, anti-patterns, and tech debt. Use this skill whenever the user says /claudboard, "onboard this project", "bootstrap Claude for this repo", "set up Claude Code for this project", "analyze this codebase", "generate rules for this project", "create CLAUDE.md", "scan this repo", "understand this project", "help me onboard to this codebase", or asks for a project health check, architecture audit, or code quality review. Also triggers on "what patterns does this repo use", "generate coding rules from this codebase", or when starting work on an unfamiliar or brownfield project. Also routes: "set up feature workflow", "install start-feature skill", "generate feature-workflow", "configure feature workflow for this repo" → claudboard-workflow. Also routes: "bootstrap workspace", "set up workspace meta-repo", "create workspace .claude repo", "share my .claude across the team", "make .claude versioned" → claudboard-workspace-init. Also routes: "join workspace", "link to workspace meta-repo", "bootstrap workspace on this machine", "clone workspace meta-repo", "connect to shared .claude" → claudboard-workspace-link.

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:

Read .claude/settings.json (if it exists).
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:

Read references/recommended-permissions.json to get the permissions.allow array.
Read .claude/settings.json (or start with {} if missing).
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.
Write back .claude/settings.json with the merged permissions.allow and the marker "_claudboard_permissions_version": "1".
Print: ✓ Recommended permissions added to .claude/settings.json
MUST NOT modify any key other than permissions.allow and _claudboard_permissions_version.

NO branch:

Write "_claudboard_permissions_version": "1-declined" into .claude/settings.json (stamp only — do not touch any other key).
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.