Health Check — act101 Agent Skill

Health Check

Depth: Level 1 (Explore).

See ../analysis-protocol/references/protocol.md for: artifact directory structure, the investigation loop, depth levels, summary format, token budget rules, and project map structure. Read that document before proceeding.

Phase 1: Parallel Tool Dispatch

Dispatch all available tools in a single parallel batch. Each subagent runs one tool, saves raw JSON to raw/<tool-name>.json, returns a structured summary.

Must-have tools:

If neither must-have tool is available, report that and stop.

Extended tools (use if available, skip and note in manifest if not):

Phase 2: Follow-up (Explore depth)

From summaries, identify the top 3-5 findings and run one targeted follow-up each:

• For each top hotspot: run skeleton to characterize what's wrong
• For untested high-coupling files: note the compound risk (untested + high blast radius)

Keep follow-ups to one round — this is a health check, not a full audit.

Trend Comparison

Score/class trend (scan history): if .act/scan-history.jsonl exists in the repo, do not hand-compare scan scores across runs — run act trends --root <repo> (add --format markdown for the report artifact). It renders the score series, per-class count deltas, top movers, and an Improving/Stable/Degrading verdict with documented bands. Encourage repos without a history file to adopt act scan --history-append in CI (full scans only — scoped runs are rejected). Count deltas are not identity tracking; the baseline mechanism below remains the identity-level security trend.

Check docs/act/ for a prior Health Check run: look for a manifest.json with "skill": "health-check". If found:

1. Load the most recent prior run's raw/hotspots.json and raw/coupling.json
2. Report deltas:
3. New hotspots since last run (files that appeared or moved up)
4. Resolved hotspots (files that disappeared or moved down)
5. Coupling changes (improving = lower instability, degrading = higher)
6. Test coverage trend (if raw/test_gaps.json exists in both runs; note whether statuses are lcov evidence-based or convention-based via the evidence field)

Security trend (scan): do not hand-diff finding lists from a prior run's raw/scan.json — the baseline mechanism computes the deltas. If the repo has a committed .act/baseline.json, pass it on the Phase 1 call: scan(root=<repo>, baseline=".act/baseline.json"). The report's baseline section IS the security trend:

• new — findings not in the baseline (regressions since it was written)
• fixed — baseline entries no longer found (remediated debt)
• baselined — acknowledged debt still present

If no baseline exists yet, record one with baseline_write (CLI: act scan --baseline-write), commit it, and the next health check gets real deltas. Score comparability is what the tools report: scores compare only across identical scan semantics — a diff-scoped scan (base_ref) is never comparable to a full-repo scan.

Duplication trend: summary.duplicated_tokens is the comparable scalar across runs — but only when both runs used the same min_tokens value. Cross-threshold comparisons (e.g. a 50-token run vs. a 100-token run) are invalid and must not be stated as a trend.

Verdict must reflect the trend: Improving / Stable / Degrading — when act trends ran, adopt its verdict unless the artifact-diff evidence contradicts it (say so explicitly if it does).

Report Structure

# Health Check: <project name>

## Health Summary
Verdict: **Improving** / **Stable** / **Degrading**.
One-paragraph assessment.
If `scan` ran, lead with its **AI-Code Health Score** (security sub-score, 0–100)
and list any `ai_config_backdoor` / `mcp_config_rce` findings as critical items.

## Duplication Snapshot
`duplicated_tokens`: N — `clone_class_count`: N — `min_tokens`: N (sourced from scan Duplication findings or direct `analyze_clones` call).

## Top Hotspots
Ranked list with skeleton context from follow-up. Severity and recommended fix.
Per-hotspot: file path, complexity score, what skeleton revealed, recommended action.

## Cohesion Issues
Low-cohesion modules with split recommendations.
Per-issue: module, cohesion score, suggested split boundary.

## Pattern Inconsistencies
Sibling files that diverge from group conventions.
Per-divergence: what's expected in the group, what's different in this file.

## Test Gaps
Untested files ranked by risk (coupling × blast radius if available, else coupling alone).
Per-gap: file path, risk factors, suggested test type.

## Trend
(Present only if prior run data exists)
| Metric | Previous | Current | Change |
|--------|----------|---------|--------|
| Top hotspot score | N | N | ↑/↓/= |
| Hotspot count (top 10) | N | N | ↑/↓/= |
| Mean instability | N | N | ↑/↓/= |
| Test gap count | N | N | ↑/↓/= |
| Duplicated tokens (same min_tokens only) | N | N | ↑/↓/= |

What improved, what degraded, what's new since last run.
If `scan` ran with a baseline, lead the security line with the `baseline`
counts: `new` = regressions, `fixed` = remediated debt, `baselined` =
acknowledged debt still present.

## Suggested Fixes
Prioritized list of specific act MCP tool calls or skills to run.

Worked Example: Duplication Snapshot

Call (when no prior scan in workflow):

analyze_clones(min_tokens=50)

Summary from act repo (2026-06-12, min_tokens 50):

{
  "summary": {
    "files_analyzed": 5392,
    "clone_class_count": 8156,
    "duplicated_tokens": 5067066,
    "min_tokens": 50,
    "test_files_excluded": 47488
  }
}

Snapshot row: duplicated_tokens: 5,067,066 — clone_class_count: 8,156 — min_tokens: 50. On the next health check, compare duplicated_tokens only if that run also uses min_tokens=50; a run at 100 produces a different floor and the delta is not comparable.

Project Map Updates

Updates the "Health Snapshot" section only. Appends to the Analysis History table.