From 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".
How this skill is triggered — by the user, by Claude, or both
Slash command
/claudboard:claudboard-refreshThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
For projects that already have `.claude/` artifacts. Compares current codebase state against existing artifacts and updates only the delta.
For projects that already have .claude/ artifacts. Compares current codebase state against existing artifacts and updates only the delta.
/refresh [path]
Path defaults to the current working directory.
Verify the project has existing Claude context — at least one of:
CLAUDE.md at project root.claude/rules/*.md.claude/skills/*/SKILL.mdIf none found: Tell the user: "No existing Claude context found. Run /analyse to scan the codebase first, then /generate to create artifacts." Stop here.
Read all existing Claude artifacts and build a coverage map:
CLAUDE.md — note which sections exist (Commands, Architecture, Rules/Skills tables, Critical Rules).claude/rules/*.mdpaths: frontmatter globs and the content summary.claude/skills/*/SKILL.mdname: and description: frontmatterreferences/ and scripts/ subdirsCheck for .claudboard/catalog.json first (primary input — all modes). If present, read umbrella_root and generated_at. All write paths use <umbrella_root>/.claude/.
Fall back to .claude/reports/claudboard-analysis.md (legacy) if catalog absent. In workspace mode without a catalog, look for cloudboard-analysis.md at the workspace root's .claude/reports/ rather than per-repo paths.
Legacy per-repo reports (claudboard-analysis-<repo>.md) are NOT read by /refresh after this change — they are superseded by the unified catalog. They remain on disk as read-only history.
Load ../claudboard/references/stack-detectors.md for detection heuristics.
If prior report exists (has generated_at timestamp):
git log --oneline --since=<generated_at> to gauge change volumegit diff --name-only <generated_at_approx_commit>..HEAD or git diff --stat to find changed areasIf no prior report:
/analyse)Compare Wide Scan results against existing artifacts:
paths: globsMonorepo service topology drift (if prior report has monorepo: true):
../claudboard/references/stack-detectors.md → "Monorepo Detection & Service Classification")cloudboard-analysis-*.md filenames):
"New service detected: {dir-name}. Run /analyse to include it.""Service removed: {dir-name}. Stale report and rules can be deleted.""Service appears renamed: {old-name} → {new-name}. Re-run /analyse to update."Workspace mode refresh behavior:
Workspace mode is detected when CWD contains subdirectories with independent .git/ repos AND each has a build file, or from catalog.mode == "workspace".
Workspace-level refresh (CWD is workspace root):
Re-run per-service surface extraction:
../claudboard/references/stack-detectors.md → "Cross-Service Surface Detection"Re-run graph construction (same as /analyse Phase 1i):
Overwrite umbrella ecosystem.md at <umbrella_root>/.claude/memories/ecosystem.md with current graph data. This is the ONLY ecosystem.md that exists — there are no per-service copies to update.
New-service detection:
/analyse to include it in the catalog and ecosystem."/analyse required.Per-service CLAUDE.md update (when catalog indicates a service's role or stack changed):
<service-dir>/CLAUDE.md — only changed sections.Service-level refresh (CWD is a single service repo):
If /refresh is run from within a service subdirectory, check for workspace context (sibling repos with .git/). If workspace context exists, warn: "Run /refresh from the workspace root to update the ecosystem graph. Service-level refresh cannot update <umbrella_root>/.claude/memories/ecosystem.md without the full workspace context." Then proceed with delta discovery for that service's rules and CLAUDE.md only.
Load ../claudboard/references/pattern-catalog.md and ../claudboard/references/quality-signals.md for assessment.
Present findings in delta format:
## Refresh Report: <repo-name>
**Last analysis:** <date from prior report, or "none">
**Changes since:** <N commits, M files changed>
### New (not covered by existing artifacts)
- <new module/directory> — needs rule or skill coverage
- <new dependency/framework> — not documented in CLAUDE.md
- <new pattern detected> — should be in rules
### Stale (artifacts reference things that changed)
- `<rule file>` — references <pattern/path> which has been modified
- `CLAUDE.md` — <section> is outdated (e.g., command changed)
- `<skill>` — references files that moved/renamed
### Missing (should exist based on codebase signals)
- No rule for <language/area> despite <N> source files
- No skill for <repetitive workflow> despite boilerplate pattern
### Up to Date (no action needed)
- `<rule>` — still accurate
- `<skill>` — still matches codebase
### Proposed Updates
**CLAUDE.md:** [specific additions/changes]
**Rules:** [new/update list with adaptive depth]
**Skills:** [new/update list]
I'll proceed now — interrupt with Esc to abort or adjust.
Apply all proposed updates. Load references as needed:
../claudboard/references/claude-md-template.md → for CLAUDE.md updates../claudboard/references/rule-templates.md → for new/updated rules../claudboard/references/skill-generation.md → for new/updated skills/claudboard-refresh SHALL NOT modify, overwrite, or delete the .claude/skills/feature-workflow/ directory or any of its contents in v1.
If .claude/skills/feature-workflow/ is detected during inventory (Phase 2c), note it in the refresh report as:
Skipped:
.claude/skills/feature-workflow/— upgrade path is opt-in via future/claudboard-workflow --upgrade
To regenerate or upgrade the feature-workflow skill, remove the directory manually and re-run /claudboard-workflow.
/generate/generate (based on quality signals)## Refresh Complete
Updated:
- CLAUDE.md [updated — added N lines]
- Rules: [list of new/updated]
- Skills: [list of new/updated]
Unchanged:
- [list of artifacts that needed no update]
Next refresh: run `/refresh` after significant codebase changes
For tech debt analysis and refactoring tickets: run `/techdebt`
Overwrite <umbrella_root>/.claude/reports/claudboard-analysis.md (all modes — single-project, monorepo, workspace) with a fresh summary combining prior data with new discoveries. Update generated_at timestamp.
Do NOT write per-repo or per-service report files. The unified umbrella report is the single source of truth.
Legacy per-repo reports (<workspace>/.claude/reports/cloudboard-analysis-<repo>.md) left by prior versions are NOT updated and NOT deleted. They remain on disk as read-only history. A future --prune-stale flag will handle cleanup.
| Condition | Behavior |
|---|---|
| Target path doesn't exist | Report error with path and stop |
No existing .claude/ context found | Tell user to run /analyse first, stop |
| Prior analysis report malformed/unreadable | Ignore it, run abbreviated Wide Scan as if no prior report |
| Git history unavailable | Skip git-based delta detection, run full Wide Scan comparison instead |
<umbrella_root>/.claude/, <umbrella_root>/CLAUDE.md, and <umbrella_root>/<service>/CLAUDE.md..claude/ writes — NEVER write to <service>/.claude/skills/, <service>/.claude/rules/, or <service>/.claude/memories/ under any mode. These per-service directories are not updated by /refresh; they simply stop receiving updates..claude/ directories are left untouched — they were produced by prior /generate versions and do no runtime harm (they were never loaded). Users may delete them at leisure. Future --prune-stale flag will automate cleanup..claude/skills/feature-workflow/ — generated by claudboard-workflow, upgrade is opt-in.| File | When to load |
|---|---|
../claudboard/references/stack-detectors.md | Step 3 — shared detection heuristics |
../claudboard/references/stack-detectors-{lang}.md | Step 3 — language-specific patterns if full Wide Scan needed |
../claudboard/references/pattern-catalog.md | Step 4 — pattern identification |
../claudboard/references/quality-signals.md | Step 4 — quality assessment |
../claudboard/references/claude-md-template.md | Step 5 — CLAUDE.md updates |
../claudboard/references/rule-templates.md | Step 5 — rule generation |
../claudboard/references/skill-generation.md | Step 5 — skill generation |
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub lukovicperisa/claudboard --plugin claudboard