coverage-trend

You are tracking Apex coverage over time so regressions are visible. The history is per-project, stored in ${CLAUDE_PLUGIN_DATA}/argo/coverage/<project>/history.jsonl (one JSON line per run).

Read Project Config First

source "${CLAUDE_PLUGIN_ROOT}/hooks/lib/config.sh"
PROJECT_NAME="$(sf_config_get '.project.name' "$ENV")"
COVERAGE_TARGET="$(sf_config_get '.quality.codeCoverageTarget' "$ENV")"
SLUG="$(printf '%s' "$PROJECT_NAME" | tr '[:upper:] /' '[:lower:]_-' | tr -cd 'a-z0-9_-')"
HISTORY_DIR="${CLAUDE_PLUGIN_DATA:-$HOME/.claude/plugin-data}/argo/coverage/${SLUG}"
mkdir -p "$HISTORY_DIR"

Input

$ARGUMENTS:

• (empty) — run sf apex run test --code-coverage and append the result to history
• record <coverage-json> — accept a coverage JSON from another tool (e.g., the existing /test-coverage skill); append it to history
• show [--last N] — print the last N runs (default 10) as a Markdown table
• diff [--vs <ref>] — compare current coverage against the run at git ref <ref> (default main HEAD)
• pr — PR mode: compute diff vs. main, exit 1 if any class regressed, exit 0 if same or improved
• --target-org <alias> / --env <name> — for the sf query in default mode

Steps

1. Run coverage (default mode) or accept input (record mode)

Default:

ORG="$(sf_config_get '.platform.defaultTargetOrg' "$ENV")"
sf apex run test --target-org "$ORG" --test-level RunLocalTests --code-coverage --result-format json --wait 30 > /tmp/cov.json

Capture per-class coverage from the JSON. Compute project-wide rollup.

2. Build the history record

{
  "ranAt":     "2026-04-28T12:30:00Z",
  "git":       { "branch": "main", "sha": "abc123", "isDirty": false },
  "overall":   { "coveredLines": 12450, "totalLines": 14200, "percent": 87.7 },
  "perClass":  [
    { "name": "OrderService", "covered": 142, "total": 150, "percent": 94.7 },
    { "name": "OrderApiClient", "covered": 38, "total": 60, "percent": 63.3 }
  ],
  "target":    85,
  "passed":    true
}

Append to ${HISTORY_DIR}/history.jsonl as a single line.

3. show [--last N]

Read the last N records. Output a Markdown table:

# Coverage Trend: <project.name>

| Run                  | Overall | Δ      | Below Target |
|----------------------|---------|--------|--------------|
| 2026-04-28T12:30:00Z | 87.7%   | +0.2%  | OrderApiClient (63.3%) |
| 2026-04-27T18:14:22Z | 87.5%   | -0.5%  | OrderApiClient (63.3%) |
| ...

4. diff [--vs <ref>]

Find the most recent history record whose git.sha matches <ref>'s HEAD. Compare class-by-class against the latest record. Output:

# Coverage Diff: HEAD vs main (abc123)

## Improved
- OrderService:    91.2% → 94.7% (+3.5%)

## Regressed
- OrderApiClient:  72.0% → 63.3% (-8.7%) ⚠️

## New classes (no baseline)
- DiscountEngine:  85.0%

## Removed classes (baseline only)
- (none)

## Overall: 87.5% → 87.7% (+0.2%)

5. pr mode

Run diff internally. Exit 1 if any class regressed by more than a configurable epsilon (default 0.5 percentage points), exit 0 otherwise. Print the diff regardless.

CI integration: emit findings (one per regression) following the standard contract:

• ruleId: "COV-REGRESSION" (warning) — class regressed but still ≥ target
• ruleId: "COV-BELOW-TARGET" (error) — class below quality.codeCoverageTarget%

6. Exit codes

• 0 — clean, or show/diff informational
• 1 — pr and any regression > epsilon, or any class below target
• 2 — query/setup error

Rules

• Per-line aggregation, not per-method. Salesforce reports lines covered/total; that's the source of truth
• Only persist on success. If sf apex run test fails, don't append a noise record
• Don't track tests; track production classes. Skip anything ending in naming.apex.testSuffix
• History is append-only. Never rewrite past lines — even if a class is renamed; the rename is its own diff event
• Truncate display. show defaults to last 10 runs to keep the table readable

Consumers

• CI pipelines run coverage-trend pr to gate merges on coverage regression
• /argo:release-notes pulls the latest record to include "Coverage: X%" in release notes
• @qa consults the history when a class drops below target to recommend which tests to add