Unit-Test Coverage Improvement (Deep)

Unit-Test Coverage Improvement (Deep)

Orchestrate paired cycles of test generation and testability refactoring. Each cycle is two subagent dispatches: first /optimus:unit-test (writes tests, measures coverage, flags untestable code), then /optimus:refactor with focus on testability (unblocks the flagged items so the next cycle can cover them). All state lives in .claude/unit-test-deep-progress.json.

Step 1: Parse Arguments and Guard Against Re-entry

Re-entry guard

If your invocation prompt body already contains HARNESS_MODE_INLINE, stop immediately with: "Deep mode cannot run inside deep mode."

Parse invocation arguments

Extract from the user's arguments:

1. --resume flag (present/absent)
2. --no-commit flag (present/absent)
3. --yes flag (present/absent) — auto-confirm the Step 3 prompt; required when invoked under claude -p or any other non-interactive session that cannot answer AskUserQuestion.
4. --max-cycles N (optional, default 5, hard cap 10)
5. Everything else → scope text (optional path filter)

Examples:

• /optimus:unit-test-deep → 5 cycles, full project
• /optimus:unit-test-deep --max-cycles 8 → 8 cycles
• /optimus:unit-test-deep src/api → scoped
• /optimus:unit-test-deep --resume → continue from existing progress file
• /optimus:unit-test-deep --no-commit → skip per-phase checkpoint commits
• claude -p "/optimus:unit-test-deep --yes src/api" → headless / CI usage; skips the Step 3 confirmation prompt

Step 2: Pre-flight Checks

Plugin root

Resolve plugin_root (the absolute path to the installed plugin) and keep it for every CLI call and subagent dispatch below — the env var does not persist across separate Bash tool calls and reads empty on some platforms (notably Windows):

1. Run echo $CLAUDE_PLUGIN_ROOT via Bash. If it is non-empty and <value>/scripts/harness_common exists (test -d), use it.
2. Otherwise derive the root from the "Base directory for this skill:" line in your invocation context — strip the trailing /skills/... segment (this skill's own directory) — and use it if <derived>/scripts/harness_common exists.
3. If neither candidate contains scripts/harness_common, stop: "Cannot resolve plugin root — ensure optimus-claude is installed via the Claude Code plugin system."

Wherever the steps below (and orchestrator-loop-*.md) write $CLAUDE_PLUGIN_ROOT, use this resolved plugin_root; if echo $CLAUDE_PLUGIN_ROOT was empty, substitute the absolute path literally.

Documentation prerequisites

Read $CLAUDE_PLUGIN_ROOT/skills/init/references/prerequisite-check.md and apply the prerequisite check. If .claude/CLAUDE.md is missing, stop: "Deep mode requires /optimus:init to set up project context first."

Test command

Read .claude/CLAUDE.md and capture the documented test command verbatim — store the exact string (e.g. npm test, pytest) as test_command. If none is documented, stop and recommend /optimus:init to set one up first (the auto-fix loop has no safety net without a test command). You pass this captured command to init in Step 4 via --test-command; init can also parse .claude/CLAUDE.md itself, but its parser is stricter than a human read, so passing the command you just read avoids a spurious "No test command found" failure on a command the CLI can't parse.

Test infrastructure

The base /optimus:unit-test skill requires a test framework to be installed. If /optimus:init flagged the test framework as missing or "installed but no tests yet," warn the user but proceed — the unit-test phase will surface the gap.

Git state

On a fresh (non---resume) run, refuse to proceed if the working tree has uncommitted changes unless --no-commit is passed.

Step 3: User Confirmation

Skip this step entirely when --resume is given, or when --yes is given (headless / CI: the caller has pre-approved the run).

Warn the user with:

Deep unit-test mode runs up to [N] cycles. Each cycle is two subagent dispatches: a unit-test phase that writes tests + measures coverage, and a refactor phase (only when the first phase flags untestable code) that unblocks testability barriers. Credit and time consumption multiplies with cycle count. Tests and refactor fixes are applied automatically without per-change approval. Press Esc twice to interrupt — state is saved per-phase; resume with /optimus:unit-test-deep --resume.

Test command: [test command]

Use AskUserQuestion — header "Deep unit-test", question "Proceed with deep unit-test?":

• Proceed — "Run the unit-test + refactor cycle loop (max [N] cycles)"
• Cancel — "Don't run deep mode"

If the user selects Cancel, stop.

Step 4: Initialize or Resume Progress

On --resume

PYTHONPATH="$CLAUDE_PLUGIN_ROOT/scripts" python -m harness_common.cli resume \
    --progress-file ".claude/unit-test-deep-progress.json" \
    [--max-cycles N] \
    --project-dir "."

Pass --max-cycles N through when the user supplied a higher cap on --resume — resume raises the persisted cycle cap (and clears a prior diminishing-returns stop) so the loop continues past the previous limit. A run that finished cleanly is archived to .done.json; --resume only continues a still-on-disk run (interrupt or diminishing-returns).

On fresh run

PYTHONPATH="$CLAUDE_PLUGIN_ROOT/scripts" python -m harness_common.cli init \
    --skill unit-test \
    --max-cycles [N] \
    --test-command "<test_command>" \
    [--scope "<scope>"] \
    [--no-commit] \
    --progress-file ".claude/unit-test-deep-progress.json" \
    --project-dir "."

Pass --no-commit through to init when the user supplied it — the mode is persisted in the progress file, so --resume keeps it without re-passing the flag (and commit-checkpoint self-skips regardless).

If init reports "progress file already exists", a prior un-archived run is on disk. Either run with --resume to continue it, or re-invoke init with --force to discard the prior progress.

Establish a baseline (fresh runs only)

Skip on --resume. On a fresh run, after init succeeds, run the baseline once to calibrate the per-cycle test timeout:

PYTHONPATH="$CLAUDE_PLUGIN_ROOT/scripts" python -m harness_common.cli baseline \
    --progress-file ".claude/unit-test-deep-progress.json" \
    --allow-red

--allow-red is passed unconditionally here: a coverage run legitimately starts with little or no passing test coverage, so a non-green baseline is not a reason to refuse. When the suite is green the command calibrates the per-cycle timeout from the measured duration (so a slow suite doesn't spuriously time out during bisection); when it isn't, it proceeds and leaves the timeout at its default.

Step 5: Run the Cycle Loop

Read $CLAUDE_PLUGIN_ROOT/references/orchestrator-loop-paired.md and follow its 11-step per-cycle body, with these parameters:

• <progress-path> = .claude/unit-test-deep-progress.json
• <max> = the cycle cap from Step 1

The paired-loop template handles:

• Dispatching the unit-test subagent with Phase: unit-test
• Recording the unit-test phase outputs (tests, coverage history, untestable items)
• Checkpoint commit for the unit-test phase
• Conditionally dispatching the refactor subagent with Phase: refactor (only when pending untestable items exist)
• Recording the refactor phase outputs (with bisection on test failure)
• Checkpoint commit for the refactor phase
• Recording cycle history + advancing the cycle counter
• Termination check (continue, convergence, cap, diminishing-returns)

Brief per-phase status updates are appropriate (e.g., "Cycle 2/5 unit-test: dispatching subagent…", "Cycle 2/5 refactor: applied 3 fixes, tests pass."). Do not narrate the subagent's findings in conversation prose — the report at Step 6 covers them.

Step 6: Final Report

PYTHONPATH="$CLAUDE_PLUGIN_ROOT/scripts" python -m harness_common.cli final-report \
    --progress-file ".claude/unit-test-deep-progress.json" \
    --archive

The report prints cycles completed, coverage baseline → final, total tests written (with file count), testability fixes applied, untestable items still pending, bugs discovered (if any), the termination reason, and git rollback guidance. The run is then archived — except on a diminishing-returns soft-exit, which the CLI leaves un-archived (prints not-archived) so it stays resumable via --resume.

Important

User approval recorded at Step 3 stands for the entire loop — tests and refactor fixes are applied without per-change confirmation. Recommend /optimus:commit next, followed by /optimus:pr once the branch is ready. Tell the user: Tip: stay in this conversation when running /optimus:commit and /optimus:pr so the implementation context is captured. Other downstream skills (/optimus:code-review, /optimus:unit-test) should still run in fresh conversations.

Tip

Unit-test-deep is the most expensive of the three orchestrator skills — each cycle is two subagent dispatches plus a full test-suite run after each. Default 5 cycles is appropriate for most codebases; raise it only when the project has substantial untestable code that takes multiple refactor rounds to unblock.