Coverage Analyzer — coverage.py XML → readable analysis

Load this skill when you need to turn a coverage.xml report into a human-readable coverage analysis: overall line/branch percent, files with zero coverage, the 10 worst-covered files, a delta vs a previously stored baseline, and a PASS/FAIL verdict against a threshold for CI.

The analyzer is pure Python 3 stdlib (xml.etree.ElementTree, json, argparse) — no dependencies, no network. It reads the exact XML format that coverage.py emits via coverage xml (Cobertura-style DTD), so it works with any tool that produces that format (pytest-cov --cov-report=xml, coverage run -m pytest && coverage xml).

The analyzer script

scripts/coverage_analyzer.py — pure Python 3 stdlib (no dependencies).

Output sections

• Total — line-rate (and branch-rate when branches were actually measured; a branch-rate="0" with branches-valid="0" is treated as "not measured", not as 0%), file count, files_with_zero_lines (files with line-rate == 0) with their names
• Worst 10 files — lowest line-rate first, ascending
• Delta vs baseline — per-file before → after → Δ table plus a total row; files absent from the baseline are marked new
• Verdict — PASS/FAIL when --threshold is given

Exit codes

Usage example (typical)

# 1. Produce the XML (coverage.py installed):
coverage run -m pytest && coverage xml

# 2. Analyze:
python3 coverage_analyzer.py --xml coverage.xml

# 3. Store a baseline on the first run:
python3 coverage_analyzer.py --xml coverage.xml --save-baseline baseline.json

# 4. On later runs, diff against the baseline and gate CI:
python3 coverage_analyzer.py --xml coverage.xml --baseline baseline.json --threshold 80

Baseline tracking workflow

1. First run — --save-baseline baseline.json writes {"files": [{"name": "...", "line_rate": 0.42}, ...], "total": 0.42}. Commit the baseline file so it is reviewable.
2. Later runs — --baseline baseline.json prints a per-file before → after → Δ table. A file that appears in the current report but not in the baseline is marked new; a file that disappeared is simply absent from the table.
3. Trend — the total row shows the overall delta in percentage points, so a regression (e.g. -5.0 pp) is visible at a glance.

Threshold gate for CI

python3 coverage_analyzer.py --xml coverage.xml --threshold 80
echo "exit=$?"   # 0 = PASS, 1 = FAIL

Use it as the last step of a test job: the script exits 1 when the total line-rate percent is below the threshold, failing the pipeline. This closes the loop after test-generator — generate tests, measure coverage, gate on the result.

Do NOT use

• If you don't have a coverage.xml — this skill only parses the coverage.py XML format; it does not run your tests or measure coverage itself. Run coverage run -m pytest && coverage xml (or pytest --cov) first.
• If you want branching visualization (branch-by-branch coverage maps, HTML reports with per-line coloring) — use coverage.py's own coverage html/coverage report or a dedicated coverage UI. This tool is a text/markdown summary + CI gate, not a visualizer.
• If your report is in a different format (lcov, JaCoCo, Cobertura from other tools) — the parser targets the coverage.py XML schema; other Cobertura-style files may parse but attribute names differ.

Canonical patterns

Full deep dive with upstream sources in references/canonical-patterns.md. Key canons:

• coverage.py XML schema (Ned Batchelder) — the line-rate/branch-rate attribute semantics this tool parses verbatim
• Cobertura DTD — the XML shape coverage.py emits (<coverage>, <packages>, <classes>, <lines>)
• pytest-cov — the --cov-report=xml pipeline that produces the input
• codecov / coveralls — the baseline-diff + threshold-gate CI model
• coverage-badge — the "percent → verdict" rendering idea (we stay text)

Files

• SKILL.md — this file
• skill.json — manifest
• scripts/coverage_analyzer.py — the stdlib analyzer (XML parse + baseline diff + threshold gate)
• references/canonical-patterns.md — coverage.py/pytest-cov/codecov/ coveralls/coverage-badge deep dive with sources

Canonical analogues

Full source depth — in references/canonical-patterns.md. Backbone:

Installation

# For opencode
cp -r skills/coverage-analyzer ~/.config/opencode/skills/

# For other agents
# Copy the skill folder to your skills directory; requires Python 3.

Note: this tool analyzes, it does not generate tests or coverage. It expects a real coverage.xml produced by coverage.py (or a compatible tool) and reports what the numbers mean — including a CI exit-code gate so coverage regressions fail the build.

Boundaries

• Do not use to generate tests or diagnose a failing test; use test-generator or systematic-debugger.