| name | diagnose |
| description | Bundle a redacted, support-grade snapshot of the current nyann state for inclusion in a bug report or support request. Combines explain-state + doctor + git config (token-redacted) + installed hook contents + nyann user-config (token-redacted) + check-prereqs into a single JSON or human-readable summary. TRIGGER when the user says "nyann is broken", "nyann diagnose", "diagnose nyann", "diagnose this repo", "support bundle", "what's wrong with nyann", "help me file a bug report", "what should I include in a nyann bug report", "/nyann:diagnose". Do NOT trigger on "is this repo healthy" — that's `doctor` (read-only audit). Do NOT trigger on "what does nyann see here" — that's `explain-state` (lighter snapshot, no health, no hook contents). Do NOT trigger on "fix this repo" — that's `retrofit`.
|
diagnose
Read-only support bundle. Aggregates everything a maintainer needs to reproduce or pinpoint a nyann bug into one paste-friendly output.
When to trigger
- User reports nyann broke on their repo and wants to file a bug
- User asks for a "support bundle", "diagnostic dump", or wants to know what to include in an issue
- User says nyann is misbehaving and wants help collecting evidence
What gets bundled
bin/diagnose.sh collects, in one structured JSON object:
nyann_version — from .claude-plugin/plugin.json
host — uname -srm, bash version, jq/git/gh versions
repo — bin/explain-state.sh --json output (stack, profile, branching, hooks, CLAUDE.md, recent commits)
health — bin/doctor.sh --json output (drift report + health score)
git — git config --list lines + git status --porcelain, with embedded URL credentials redacted via nyann::redact_url
hook_files — first 8 KB of .git/hooks/pre-commit, .git/hooks/commit-msg, .husky/pre-commit, .husky/commit-msg, .pre-commit-config.yaml (each null when absent)
nyann_config — ~/.claude/nyann/preferences.json + team_profile_sources[] with URLs redacted
prereqs — bin/check-prereqs.sh --json
Every field flows through redaction. The bundle is safe to paste into a public GitHub issue.
Execution flow
Phase 1: Explain what diagnose collects
Tell the user: "I'll run bin/diagnose.sh against this repo. It collects state, drift, hooks, and configuration into one bundle. URL credentials and tokens are redacted before output."
Phase 2: Run the bundle
Run bin/diagnose.sh --target . --json (default mode is human-readable; --json for the paste-into-issue version).
If the user wants to share publicly, suggest:
- Pipe to a file:
bin/diagnose.sh --target . --json > nyann-diagnose-$(date +%Y%m%d).json
- OR use the human-readable mode (
bin/diagnose.sh --target .) for a quick eyeball summary
Phase 3: Surface the obvious problems
Before handing the bundle off, look at:
health.summary.broken_links > 0 → user has broken internal links
health.summary.missing > 0 → user has missing hygiene files
repo.claude_md.bytes > 8192 → CLAUDE.md over the hard cap
prereqs[].status of missing for any kind: hard → nyann lacks a required tool
Mention the obvious ones; the maintainer will look at the bundle for the rest.
Phase 4: Where to file
Direct the user to the right surface:
Key constraints
- Read-only. Mirrors doctor's contract: never writes, never persists, never opens a network call beyond what
doctor --json already does.
- Redaction is the contract. Any URL with embedded credentials passes through
nyann::redact_url before emission. Tokens never reach stdout.
- Truncated hook content (8 KB cap) so a multi-MB hook script doesn't bloat the bundle.
- Per-tier degradation: if
gh is missing, the bundle still emits with host.gh = "missing". If ~/.claude/nyann/preferences.json doesn't exist, nyann_config = null. If the target isn't a git repo, git, health, and repo are stub-empty but the bundle still emits.
When to hand off
- "Fix the issues" / "remediate" →
retrofit skill (targeted fixes)
or doctor skill (audit first, then decide).
- "Just check health, not a full bundle" →
doctor skill.
- "What does nyann see here?" →
explain-state skill (lighter, no
health or hook contents).