Shell exit evidence
std/command.shell_exit_evidence(source, shell) reports what a zero process
exit establishes about commands visible in a shell line. It never executes or
rewrites the command and never supplies constituent exit codes.
The result has two closed fields:
| Field | Values |
|---|---|
scope | all_commands, masked, unmeasured |
reason | all_commands, syntax, dialect, chain_status, pipeline_status, compound_command, nested_execution, shell_state |
all_commands covers a plain POSIX command and plain commands joined with
&&. A pipeline also requires an earlier successful set -o pipefail. Bash
and Zsh additionally admit a leading set -o pipefail; prefix. A literal shell
script file is treated as one external program.
masked means a pipeline, fallback, separator or background operation can
hide a constituent failure. unmeasured means the grammar, dialect, nested
execution, control flow or shell-state change prevents this proof. Missing
evidence must not be treated as all_commands.
The same proof appears in
command_risk_scan(...).execution_semantics.exit_evidence, using the request's
resolved execution mode and shell. An argv request preserves its argument
boundaries. It is not reinterpreted as an unquoted shell string.
This contract covers command-line structure. It does not certify a program's internal checks, imported shell functions, test discovery, collected file coverage or the correctness of a test assertion. A zero exit and a test count alone cannot establish that a newly added test file was collected.