CLI reference

The complete surface of shipmoor test-evidence in v0.8.0: a standalone top-level command, a peer to scan, review, and claim-check, never a flag on any of them. It tells you whether the tests that ran on a change actually ran and passed, whether a failing test quietly disappeared, and whether a kept test was gutted until it couldn’t fail.

usage: shipmoor test-evidence [-h] [--staged] [--diff DIFF] [--all]
                               [--from FROM_REF] [--to TO_REF]
                               [--agent AGENT] [--high-assurance]
                               [--attest-out ATTEST_OUT] [--json]
                               [--terse] [--no-color]
                               [target]

shipmoor test-evidence requires the test_evidence entitlement (IC tier, the same tier as claim_check today; see Plans & tiers). The one exception is the hook subcommand below: it is never entitlement-gated, on any tier, because the spool it fills has to exist before anyone can ask whether the entitlement is present.

Positional argument

Selectors: what gets checked

By default, shipmoor test-evidence resolves the same change surface claim-check does: your working tree (tracked changes plus untracked files) as a diff. The flags below narrow or widen that.

shipmoor test-evidence .                               # the working tree (default)
shipmoor test-evidence . --staged                      # the staged index
shipmoor test-evidence . --diff main...HEAD            # a raw diff spec
shipmoor test-evidence . --from origin/main --to HEAD  # a committed range
shipmoor test-evidence . --all                         # the whole tree

--staged, --diff, and --all are mutually exclusive with each other: only one scope selector applies at a time. --from and --to must be given together, and combining that pair with --staged, --diff, or --all is rejected as an invalid combination.

The weakened-test judge: --agent

passed and reduced are fully deterministic and never touch an agent, whether or not --agent is set. Omit it and weakened simply abstains: it discloses not judged, never a guess.

shipmoor test-evidence . --agent claude

See Signals & anti-gaming for why only weakened needs a judge at all, and BYO-Judge for the wire contract these presets share with Claim Check’s own judge.

Spot verification: --high-assurance

Honest limitation. The subset re-runner this flag describes is not built yet. Until it is, --high-assurance does not silently trust an unverified harvest as if it had been spot-checked: it conservatively downgrades that harvest and discloses it as not_checked instead. Passing the flag today gets you a more cautious disclosure, not an actual re-run. That will change once the re-runner ships.

Output flags

shipmoor test-evidence . --attest-out .shipmoor/test-evidence.attest.json --json

The hook subcommand

shipmoor test-evidence hook [-h] [--dir DIR] [--config CONFIG]

hook is a lifecycle hook, not a report. It is never entitlement-gated; it always runs, for every tier. Wire it as a Claude Code PostToolUse hook on the Bash tool (project or user level .claude/settings.json), and the harness pipes each tool-result event to its stdin:

{
  "hooks": {
    "PostToolUse": [
      { "matcher": "Bash", "hooks": [{ "type": "command", "command": "shipmoor test-evidence hook" }] }
    ]
  }
}

When the piped event looks like a real test invocation, hook records it to a local evidence spool at .shipmoor/test-evidence/spool/, bound to the working-tree hash at that moment, for a later shipmoor test-evidence run to harvest. hook always exits 0: a lifecycle hook must never interrupt an agent’s turn, so a non-test command, or a working directory that isn’t a repo, is a silent no-op, never an error.

Config

Defaults come from a test_evidence: block in .shipmoor.yaml. Precedence is flags, then config, then built-in defaults. Every key is optional; the built-ins already cover Shipmoor’s four supported languages (Python, TypeScript, JavaScript, Go). Override any of them explicitly, for example to point at a non-standard report location:

test_evidence:
  test_globs: ["**/test_*.py", "**/*_test.go", "**/*.test.ts", "**/*.spec.ts"]
  report_globs: ["junit.xml", "coverage/lcov.info", ".shipmoor/test-evidence/*.xml"]
  test_command_patterns: ["pytest*", "go test*", "npm test*", "jest*"]
  high_assurance: false

Exit codes

shipmoor test-evidence follows the same stable exit-code contract as the rest of the CLI:

A passed result of not_checked, or a weakened result of not judged, is a disclosure, not a finding: on its own it does not flip the exit code to 1. Only a disclosed reduced or weakened finding does that. See Output formats & exit codes for the same contract elsewhere in the CLI.

Example invocations

# Standalone run over the working tree, with the weakened-test judge:
shipmoor test-evidence . --agent claude --json --attest-out .shipmoor/test-evidence.attest.json

# In CI, over a committed range:
shipmoor test-evidence . --from origin/main --to HEAD --terse

Next

• Test Evidence overview: what the three signals are and why harvesting beats re-running.
• Quickstart: wire the hook and run your first harvest.
• Signals & anti-gaming: the full depth on passed, reduced, and weakened.
• Using Test Evidence with Claim Check: how these signals compose into the bigger gate.
• Claim Check CLI reference: the peer command’s full surface.