health

If this session was compacted, the copy of this skill in context may be cut off. Read ${CLAUDE_SKILL_DIR}/SKILL.md from disk and resume at the step you are on.

When to invoke

Use when: "health check", "code quality", "how healthy is the codebase", "run all checks", "quality score".

Preamble

eval "$("${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-slug" 2>/dev/null)" 2>/dev/null || SLUG="unknown"
_LEARN_FILE="${VIBESTACK_HOME:-$HOME/.vibestack}/projects/${SLUG:-unknown}/learnings.jsonl"
if [ -f "$_LEARN_FILE" ]; then
  _LEARN_COUNT=$(wc -l < "$_LEARN_FILE" 2>/dev/null | tr -d ' ')
  echo "LEARNINGS: $_LEARN_COUNT entries loaded"
  if [ "$_LEARN_COUNT" -gt 5 ] 2>/dev/null; then
    "${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-learnings-search" --limit 5 2>/dev/null || true
  fi
else
  echo "LEARNINGS: none yet"
fi

{{include lib/snippets/session-host.md}}

{{include lib/snippets/decision-brief.md}}

{{include lib/snippets/working-protocols.md}}

{{include lib/snippets/state-protocols.md}}

User-invocable

When the user types /health, run this skill. /health skills runs Target: skills instead of Steps 1-6. /health --fix runs Mode: --fix instead of Steps 2-6.

Step 1: Detect Health Stack

Read CLAUDE.md and look for a ## Health Stack section. If found, parse the tools listed there and skip auto-detection. Then print `MISSING: <category> (not listed in

Health Stack)` for each of the five categories (typecheck, lint, test, deadcode,

shell) the section omits.

If no ## Health Stack section exists, auto-detect available tools:

# A Node tool installed as a devDependency is not on PATH; run it from node_modules/.bin.
node_tool() { if [ -x "node_modules/.bin/${1}" ]; then echo "./node_modules/.bin/${1}"; else echo "${1}"; fi; }
setopt +o nomatch 2>/dev/null || true
health_found=$(
# Type checker
[ -f tsconfig.json ] && echo "TYPECHECK: $(node_tool tsc) --noEmit"

# Linter
[ -f biome.json ] || [ -f biome.jsonc ] && echo "LINT: $(node_tool biome) check ."
ls eslint.config.* .eslintrc.* .eslintrc 2>/dev/null | sed -n '1p' | xargs -I{} echo "LINT: $(node_tool eslint) ."
[ -f .pylintrc ] || [ -f pyproject.toml ] && grep -q "pylint\|ruff" pyproject.toml 2>/dev/null && echo "LINT: ruff check ."
command -v terraform >/dev/null 2>&1 && [ -n "$(find . -maxdepth 3 -name '*.tf' -not -path '*/.terraform/*' 2>/dev/null | sed -n '1p')" ] && echo "LINT: terraform fmt -check -recursive"
command -v cfn-lint >/dev/null 2>&1 && cfn_files=$(find . -maxdepth 3 \( -name node_modules -o -name .git \) -prune -o -type f \( -name '*.yaml' -o -name '*.yml' -o -name '*.json' -o -name '*.template' \) -exec grep -l 'AWSTemplateFormatVersion' {} + 2>/dev/null | sort | tr '\n' ' ') && [ -n "$cfn_files" ] && echo "LINT: cfn-lint ${cfn_files% }"
command -v actionlint >/dev/null 2>&1 && [ -d .github/workflows ] && echo "LINT: actionlint"

# Test runner: only a non-empty, single-line scripts.test is a command.
[ -f package.json ] && python3 -I -c '
import json, sys
try:
    t = json.load(open(sys.argv[1])).get("scripts", {}).get("test")
except Exception:
    sys.exit(0)
if isinstance(t, str) and t.strip():
    t = t.strip()
    print("TEST: none (multi-line test script; add it to a ## Health Stack section)" if "\n" in t else "TEST: " + t)
' package.json
[ -f pyproject.toml ] && grep -q "pytest" pyproject.toml 2>/dev/null && echo "TEST: pytest"
[ -f Cargo.toml ] && echo "TEST: cargo test"
[ -f go.mod ] && echo "TEST: go test ./..."

# Dead code
command -v knip >/dev/null 2>&1 && echo "DEADCODE: knip"
[ -f package.json ] && grep -q '"knip"' package.json 2>/dev/null && echo "DEADCODE: npx knip"

# Shell linting
command -v shellcheck >/dev/null 2>&1 && ls *.sh scripts/*.sh bin/*.sh 2>/dev/null | sed -n '1p' | xargs -I{} echo "SHELL: shellcheck"
)
[ -n "$health_found" ] && printf '%s\n' "$health_found"

# One MISSING line, with its reason, per category nothing was detected for.
health_missing() { printf 'MISSING: %s (%s)\n' "${1}" "${2}"; }
printf '%s\n' "$health_found" | grep -q '^TYPECHECK: ' || health_missing typecheck "no tsconfig.json"
printf '%s\n' "$health_found" | grep -q '^LINT: ' || health_missing lint "no biome, eslint or ruff config; no terraform, cfn-lint or actionlint target with its binary on PATH"
if ! printf '%s\n' "$health_found" | grep '^TEST: ' | grep -vq '^TEST: none ('; then
  test_reason=$(printf '%s\n' "$health_found" | sed -n 's/^TEST: none (\(.*\))$/\1/p' | sed -n '1p')
  health_missing test "${test_reason:-no scripts.test, pytest, Cargo.toml or go.mod}"
fi
printf '%s\n' "$health_found" | grep -q '^DEADCODE: ' || health_missing deadcode "knip neither on PATH nor in package.json"
printf '%s\n' "$health_found" | grep -q '^SHELL: ' || health_missing shell "shellcheck not on PATH, or no *.sh in ., scripts/ or bin/"

A TEST: none (...) line is not a command: the test category is missing, for the reason in parentheses. Every MISSING: category is unavailable (null in Step 3) and its reason goes into the dashboard's Unavailable line.

Use Glob to search for shell scripts:

• **/*.sh (shell scripts in the repo)

After auto-detection, present the detected tools via AskUserQuestion:

"I detected these health check tools for this project:

• Type check: tsc --noEmit
• Lint: biome check .
• Tests: bun test
• Dead code: knip
• Shell lint: shellcheck *.sh

A) Looks right -- persist to CLAUDE.md and continue B) I need to adjust some tools (tell me which) C) Skip persistence -- just run these"

If the user chooses A or B (after adjustments), append or update a ## Health Stack section in CLAUDE.md:

## Health Stack

- typecheck: tsc --noEmit
- lint: biome check .
- test: bun test
- deadcode: knip
- shell: shellcheck *.sh scripts/*.sh

Step 2: Run Tools

Run each detected tool. For each tool:

1. Record the start time
2. Run the command, capturing complete stdout and stderr in a private temporary log
3. Record the checker's own exit code, before any parser or display command runs
4. Record the end time
5. Count findings from the complete log, then show its last 50 lines for the report

# Capture one tool. Per tool, set health_tool, the command line and the count pattern.
(
  umask 077
  health_tool=typecheck
  health_pattern='error TS'
  health_capture_error() {
    printf 'TOOL:%s ERROR:capture-%s\n' "$health_tool" "${1}"
    exit 125
  }
  health_log=$(mktemp "${TMPDIR:-/tmp}/vibe-health.XXXXXX") || health_capture_error log
  trap 'rm -f -- "$health_log"' EXIT
  health_start=$(date +%s) || health_capture_error timing
  # Open the log first, so a redirection failure cannot pass for a checker result.
  exec 3>"$health_log" || health_capture_error redirection
  if tsc --noEmit >&3 2>&1; then
    health_status=0
  else
    health_status=$?
  fi
  exec 3>&-
  health_end=$(date +%s) || health_capture_error timing
  # awk prints 0 for no matches, so an empty log still yields a count.
  health_count=$(awk -v p="$health_pattern" 'index($(0), p) { n++ } END { print n+0 }' "$health_log") || health_capture_error parsing
  tail -50 "$health_log" || health_capture_error display
  printf 'TOOL:%s EXIT:%s DURATION:%ss COUNT:%s\n' "$health_tool" "$health_status" "$((health_end-health_start))" "$health_count"
)

EXIT is the checker's status, never that of tail, awk or a pipe. COUNT comes from the whole log, not the 50 lines shown. Swap in the tool's command and the fixed string that marks one finding (error TS for tsc; Step 3 lists the others); for a tool whose summary line carries the count, read it from the same log.

Run tools sequentially, one block per tool (some share resources or lock files). A failing checker does not stop the later ones.

SKIPPED, FAILED or ERROR — decide before and after running:

• SKIPPED — decided before running: the tool's binary is absent (command -v <binary> finds nothing, no local node_modules/.bin entry) and no ## Health Stack entry names it. Record the reason. Only a skip redistributes weight. A binary that exists only in node_modules/.bin is present: run it as ./node_modules/.bin/<binary>, including when a ## Health Stack entry names it bare.
• FAILED — the command ran and could not execute the check: exit 126 or 127 (typo, missing binary in a configured ## Health Stack command, no permission). It scores 0. A configured command that does not run is a broken check, not a missing one, and must never raise the score.
• ERROR — the capture itself broke (TOOL:<name> ERROR:capture-...: log, redirection, timing, parsing or display). The category gets no score, the composite is N/A — capture failed, and no history row is written.

Step 3: Score Each Category

Score each category on a 0-10 scale using this rubric:

Parsing tool output for counts: use the complete captured log (COUNT above), never the displayed tail. A non-zero exit is never CLEAN and never 10. Exit 126 or 127 is FAILED and scores 0, whatever the log says. Any other non-zero exit whose full log yields no count scores 4 (as for a test runner that reports only its exit code); keep its output for the details section.

• tsc: Count lines matching error TS in output.
• biome/eslint/ruff: Count lines matching error/warning patterns. Parse the summary line if available.
• Tests: Parse pass/fail counts from the test runner output. If the runner only reports exit code, use: exit 0 = 10, any other non-zero exit except 126/127 = 4 (assume some failures).
• knip: Count lines reporting unused exports, files, or dependencies.
• shellcheck: Count distinct findings (lines starting with "In ... line").

Composite score: compute it in code, never by hand. Pass every category once as name=<score>, where the score is an integer 0-10, null for SKIPPED, or error for a capture ERROR. Skipped weight is redistributed proportionally among the scored categories; the weights are typecheck 25%, lint 20%, test 30%, deadcode 15%, shell 10%.

python3 -I -c '
import sys
w = {"typecheck": 0.25, "lint": 0.20, "test": 0.30, "deadcode": 0.15, "shell": 0.10}
s = {}
for a in sys.argv[1:]:
    k, sep, v = a.partition("=")
    if not sep or k not in w or k in s:
        sys.exit("bad argument: " + a)
    if v not in ("null", "error") and not (v.isdigit() and int(v) <= 10):
        sys.exit("bad score for " + k + ": " + v)
    s[k] = v
if set(s) != set(w):
    sys.exit("pass every category: " + " ".join(w))
scored = [k for k in w if s[k] not in ("null", "error")]
errors = [k for k in w if s[k] == "error"]
print("CHECKED: " + (" ".join(scored) or "none"))
print("UNAVAILABLE: " + (" ".join(k for k in w if s[k] == "null") or "none"))
print("COVERAGE: %d/%d" % (len(scored), len(w)))
if errors:
    print("COMPOSITE: N/A - capture failed (" + " ".join(errors) + ")")
elif not scored:
    print("COMPOSITE: N/A - no checks ran")
else:
    c = sum(int(s[k]) * w[k] for k in scored) / sum(w[k] for k in scored)
    print("COMPOSITE: %.1f" % c + ("" if len(scored) == len(w) else " (partial coverage)"))
' typecheck=10 lint=8 test=10 deadcode=7 shell=null

Replace the example values with this run's results. Report the COMPOSITE line as printed; a numeric composite exists only when it prints a number.

Step 4: Present Dashboard

Present results as a clear table:

CODE HEALTH DASHBOARD
=====================

Project: <project name>
Branch:  <current branch>
Date:    <today>

Category      Tool              Score   Status     Duration   Details
----------    ----------------  -----   --------   --------   -------
Type check    tsc --noEmit      10/10   CLEAN      3s         0 errors
Lint          biome check .      8/10   WARNING    2s         3 warnings
Tests         bun test          10/10   CLEAN      12s        47/47 passed
Dead code     knip               7/10   WARNING    5s         4 unused exports
Shell lint    shellcheck        10/10   CLEAN      1s         0 issues

COMPOSITE SCORE: 9.2 / 10
Coverage: 5/5 categories checked
Checked: typecheck, lint, test, deadcode, shell
Unavailable: none

Duration: 23s total
VERDICT: NOT READY (lint, deadcode exited non-zero)

Use these status labels:

• 10: CLEAN
• 7-9: WARNING
• 4-6: NEEDS WORK
• 0-3: CRITICAL
• Command could not run (exit 126/127): FAILED (scores 0)
• Tool not available: SKIPPED (no score)
• Capture broke: ERROR (no score; composite is N/A)

Verdict line. End the dashboard with one of:

• VERDICT: READY only when every required check ran and exited 0 and all five categories were checked. The required checks are every ## Health Stack entry, or every detected tool when the stack was auto-detected.
• VERDICT: NOT READY when every required check ran and at least one exited non-zero.
• VERDICT: INCOMPLETE when any required check was SKIPPED, FAILED or ERROR, or fewer than five categories were checked. Name each missing one. A high composite does not change this: never READY with partial coverage.

Always show the coverage lines. With skipped categories, label the score — partial coverage (e.g. COMPOSITE SCORE: 8.0 / 10 — partial coverage, Coverage: 2/5 categories checked) and list each unavailable category with the reason from its Step 1 MISSING: line (or its SKIPPED reason), e.g. Unavailable: deadcode (knip neither on PATH nor in package.json). When no check ran, show COMPOSITE SCORE: N/A — no checks ran and say which tools need installing or configuring; never show 10/10 for an empty run.

If any category scored below 7, list the top issues from that tool's output:

DETAILS: Lint (3 warnings)
  biome check . output:
    src/utils.ts:42 — lint/complexity/noForEach: Prefer for...of
    src/api.ts:18 — lint/style/useConst: Use const instead of let
    src/api.ts:55 — lint/suspicious/noExplicitAny: Unexpected any

Step 5: Persist to Health History

eval "$("${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-slug" 2>/dev/null)" && mkdir -p "${VIBESTACK_HOME:-$HOME/.vibestack}/projects/$SLUG"
echo "PROJECT_DIR: ${VIBESTACK_HOME:-$HOME/.vibestack}/projects/$SLUG"

Only when the composite is numeric, append one JSONL line to <PROJECT_DIR>/health-history.jsonl (where <PROJECT_DIR> is the path the block above printed on its PROJECT_DIR: line). A run with no checks or a capture ERROR writes nothing and leaves the existing history as it is:

{"ts":"2026-03-31T14:30:00Z","branch":"main","score":9.2,"typecheck":10,"lint":8,"test":10,"deadcode":7,"shell":10,"duration_s":23}

Fields:

• ts -- ISO 8601 timestamp
• branch -- current git branch
• score -- composite score (one decimal)
• typecheck, lint, test, deadcode, shell -- individual category scores (integer 0-10)
• duration_s -- total time for all tools in seconds

If a category was skipped, set its value to null.

Step 6: Trend Analysis + Recommendations

Read the last 10 entries from <PROJECT_DIR>/health-history.jsonl (if the file exists and has prior entries).

eval "$("${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-slug" 2>/dev/null)" && mkdir -p "${VIBESTACK_HOME:-$HOME/.vibestack}/projects/$SLUG"
tail -10 "${VIBESTACK_HOME:-$HOME/.vibestack}/projects/$SLUG/health-history.jsonl" 2>/dev/null || echo "NO_HISTORY"

Compare like-for-like coverage only. For each history row, the scored set is the categories with a non-null value (a missing field counts as null). Report a delta, IMPROVING or REGRESSIONS DETECTED only against rows whose scored set equals this run's exactly. If the previous run's set differs, say Coverage changed — scores are not comparable and label nothing an improvement or a regression; older rows may still be listed with their unavailable categories marked. An N/A run has no trend.

If comparable prior entries exist, show the trend:

HEALTH TREND (last 5 runs)
==========================
Date          Branch         Score   TC   Lint  Test  Dead  Shell
----------    -----------    -----   --   ----  ----  ----  -----
2026-03-28    main           9.4     10   9     10    8     10
2026-03-29    feat/auth      8.8     10   7     10    7     10
2026-03-30    feat/auth      8.2     10   6     9     7     10
2026-03-31    feat/auth      9.2     10   8     10    7     10

Trend: IMPROVING (+1.0 since last run)

If score dropped vs the previous run with identical coverage:

1. Identify WHICH categories declined
2. Show the delta for each declining category
3. Correlate with tool output -- what specific errors/warnings appeared?

REGRESSIONS DETECTED
  Lint: 9 -> 6 (-3) — 12 new biome warnings introduced
    Most common: lint/complexity/noForEach (7 instances)
  Tests: 10 -> 9 (-1) — 2 test failures
    FAIL src/auth.test.ts > should validate token expiry
    FAIL src/auth.test.ts > should reject malformed JWT

Health improvement suggestions (always show these):

Prioritize suggestions by impact (weight * score deficit):

RECOMMENDATIONS (by impact)
============================
1. [HIGH]  Fix 2 failing tests (Tests: 9/10, weight 30%)
   Run: bun test --verbose to see failures
2. [MED]   Address 12 lint warnings (Lint: 6/10, weight 20%)
   Run: biome check . --write to auto-fix
3. [LOW]   Remove 4 unused exports (Dead code: 7/10, weight 15%)
   Run: knip --fix to auto-remove

Rank by weight * (10 - score) descending. Only show categories below 10.

Mode: --fix

/health --fix runs the checks Step 1 found, fixes the source until they pass, and proves the result from the evidence ledger. It writes no score and no history row.

Checks. Run Step 1 (a ## Health Stack section, or auto-detection). Each detected command is one check, labelled by its category (typecheck, lint, test, deadcode, shell; lint-2 for a second command in one category). With no check at all, print CHECKS: none and stop: there is nothing to make green, and saying it passed would be a lie.

Consent. The commands come from the repo and run its code. Interactive: print every command, and before the first run in a repo with no earlier health record ("${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-evidence" read --skill health --limit 1 prints no evidence recorded) ask with AskUserQuestion: A) run them, B) stop. SESSION_KIND: spawned or headless: run only when invoked as /health --fix --yes; otherwise print FIX: needs --yes in a spawned or headless session and stop.

Run the checks. One block per run; one fix_check line per check, with its label and its exact command as ONE single-quoted string:

_EV=${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-evidence
_FIX_LOG=$(mktemp -d "${TMPDIR:-/tmp}/vibe-health-fix-XXXXXXXX") || { echo "FIX: cannot create a log directory"; exit 125; }
fix_check() {
  "$_EV" run --skill health --label "${1}" -- "${2}" > "$_FIX_LOG/${1}.txt" 2>&1
  _rc=$?
  printf 'CHECK: %s exit=%s log=%s\n' "${1}" "$_rc" "$_FIX_LOG/${1}.txt"
  if [ "$_rc" -ne 0 ]; then tail -50 "$_FIX_LOG/${1}.txt"; fi
}
fix_check '<check>' '<command>'

If a log says vibe-evidence itself was not found, stop: this mode proves its result from the ledger, and without it there is nothing to prove.

Rules for every run:

• Only exit 0 is green. Any other exit=, a missing line or "mostly passed" is red, whatever the log says.
• Pre-existing means red on the first run this session, before any edit. A check that turns red after your edit is yours, including a test you never touched that your changed source now breaks.
• ENV, TIMEOUT or a missing secret is reported with the log line that shows it. Never stub, mock or skip around it.
• Never edit a test, a checker config or a script (package.json scripts, lint rules, CI files, ## Health Stack) to get green. Fix the source the check exercises; a wrong test is reported, not rewritten.
• Stop after 5 runs, or on the same failure twice after an attempted fix. Report what is still red and hand it to /investigate.

Finish from the ledger. Do not re-run anything; ask the ledger whether each check's exact command passed at the tree as it is now. One fix_ready line per check, same label and command as the last run:

_EV=${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-evidence
TREE_NOW=$("${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-tree-hash" 2>/dev/null || echo unknown)
_fix_bad=""
fix_ready() { "$_EV" check --skill health --label "${1}" --expect-cmd "${2}" || _fix_bad="$_fix_bad ${1}"; }
fix_ready '<check>' '<command>'
if [ -z "$_fix_bad" ]; then echo "READY at tree $TREE_NOW"; else echo "NOT READY:$_fix_bad"; fi

Report the READY or NOT READY line as printed, with each check's FRESH or STALE line, the pre-existing failures, and the files you changed. An edit after the last run makes it NOT READY: run the checks again rather than claiming it.

Target: skills

/health skills reports on the installed skills instead of the codebase: how often each one is used, what its description costs every session, and what to do about it. Both tools are read-only.

"${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-skill-usage" --days 90
"${VIBESTACK_HOME:-$HOME/.vibestack}/bin/vibe-context-budget" --runtime claude --rendered "$HOME/.claude/skills" --config-dir "" --top 0 --warn-only

vibe-skill-usage counts typed (the user typed /<name>) and model (the model called the skill) from Claude Code transcripts, nested for the share found in subagent transcripts, and tracked from the slash-command analytics file. Typed and tracked overlap — never add them. vibe-context-budget gives each skill's description length (--top 0 lists every skill; --config-dir "" skips the repo config payload, which an installed copy does not have).

Present one row per installed skill:

SKILL HEALTH (last 90 days, claude measured; cursor, kiro, codex, copilot unmeasured)
Skill           Typed  Model  Last seen    Desc chars  Proposal
--------------  -----  -----  -----------  ----------  ------------------------------
ship               12     30  2026-10-09          112  Keep — used weekly, both ways
diagram             0      0  -                   140  Improve — never auto-invoked; description names no trigger phrase
old-helper          0      0  -                   210  Retire — no use in 90 days on the one measured runtime and covered by /vibe

Propose exactly one of Keep, Improve, Retire or Merge into <skill> per skill, with a reason a reader can check without this conversation (the counts, the overlapping skill, the description problem).

Rules:

• Unmeasured is not zero. Usage on Cursor, Kiro, Codex or Copilot CLI, and in transcripts the tool could not read, is unmeasured, and unmeasured never justifies Retire. Say which runtimes the user works in before proposing Retire; a skill with zero Claude use may be in daily use elsewhere.
• Retire is a proposal. This target never deletes, moves or edits a skill. Point to /vibe-gc for the removal itself.
• Merge into names the surviving skill and what moves into it.
• Improve names the change: a description that never matches how users ask, a description long enough to dominate the budget, a trigger list that collides with another skill.

{{include lib/snippets/capture-learnings.md}} The score itself belongs in health-history.jsonl, not here. What belongs here is what you had to work out to produce it: a tool that needs a flag to be parseable, a category whose count is misleading on this repo, a check that is slow enough to be worth skipping on a quick pass.

Important Rules

1. Wrap, don't replace. Run the project's own tools. Never substitute your own analysis for what the tool reports.
2. Read-only outside --fix. Never fix issues unless the user ran /health --fix. Present the dashboard and let the user decide.
3. Respect CLAUDE.md. If ## Health Stack is configured, use those exact commands. Do not second-guess.
4. Skipped is not failed. Skip only a tool whose absence was established before running, show coverage, and redistribute weight among scored categories only. A command that ran and failed — including exit 127 — is FAILED, never SKIPPED.
5. Show raw output for failures. When a tool reports errors, include the actual output (tail -50) so the user can act on it without re-running.
6. Trends require comparable history. On the first scored run, say "First health check -- no trend data yet. Run /health again after making changes to track progress." A changed coverage set or an N/A run has no score delta.
7. Be honest about scores. A codebase with 100 type errors and all tests passing is not healthy. The composite score should reflect reality.