How It Works

The problem

Coverage tools tell you what percentage of your code is tested and which lines are missing. But they stop there.

$ pytest --cov=myapp
Name                 Stmts   Miss  Cover
----------------------------------------
myapp/validator.py      47     12    74%

74% coverage. 12 lines missing. But which 12 lines? And what tests would actually cover them?

The solution

code-covered bridges the gap between “lines not covered” and “tests you should write.” It reads your coverage.json, walks the AST to understand what each uncovered block actually does, and generates prioritized test suggestions with ready-to-use stubs.

$ code-covered coverage.json

============================================================
code-covered
============================================================
Coverage: 74.5% (35/47 lines)
Files analyzed: 1 (1 with gaps)

Missing tests: 4
  [!!] CRITICAL: 2
  [!]  HIGH: 2

Top suggestions:
  1. [!!] test_validator_validate_input_handles_exception
       In validate_input() lines 23-27 - when ValueError is raised

  2. [!!] test_validator_parse_data_raises_error
       In parse_data() lines 45-45 - raise ParseError

The five-step pipeline

1. Parse coverage.json — Reads the JSON report produced by pytest-cov.

2. AST analysis — Parses each source file with Python’s stdlib ast module to map every uncovered line back to its function, class, and enclosing block.

3. Context detection — Classifies what each uncovered block does:

   • Is it an exception handler (except block)?
   • Is it a conditional branch (if/elif/else)?
   • Is it a loop body (for/while)?
   • Is it a raise statement?

4. Template generation — Creates specific, context-aware test templates with Arrange/Act/Assert scaffolding and setup hints (e.g., “Mock HTTP requests with responses or httpx”).

5. Prioritization — Ranks suggestions by risk. Error paths and raise statements surface first, then branches, then loops, then everything else.

Zero dependencies

code-covered uses only the Python standard library. No heavy installs, no API calls, no runtime magic. It works anywhere pytest-cov produces a coverage.json.