argus 0.5.0-rc.2

argus 0.5.0-rc.2

The second release candidate. It is a prerelease: install.sh never resolves
it as latest, and the Homebrew formula is not bumped to it. It exists so the
new refusals below can be tried on a Linux machine before 0.5.0 is tagged, and
it is the first release built with the updated QEMU and Buildx actions.

Since 0.5.0-rc.1:

• dast speaks as dast:, its own voice. Its stages are section headings, and
  where the evidence is is said once, as dast: output. The Guard's rows inside
  a run are unchanged.
• dast --intrusive is refused in one block under the rule tier, exit 1 as
  before; a run into argus's own inputs, or into a Project outside .argus/, is
  refused under argus-inputs, argus-data or run-dir.
• A committed output: in .argus.yml must be valid UTF-8 with no character that
  can rewrite a terminal line; a hostile value could hide rows or forge one.
• A directory named with non-ASCII letters, such as café, is printed as itself
  in a private-out refusal, with its chmod line to paste.
• No command formats its own status line; a test keeps it that way.

Breaking, read first:

• A committed profile: in .argus.yml must be a profile directory name:
  lowercase letters, digits, dot, underscore and hyphen, not starting with a
  dot. A config naming "Generic" or "acme app" is now refused under the rule
  project-config, and init asks again rather than writing one.
• argus dast and argus guard run refuse an --out that another user could
  replace while a scanner starts (rule private-out): a directory on the way
  that belongs to another user, or that others can write to without the sticky
  bit. A directory writable only by your own per-user group, as umask 002
  leaves one on most Linux systems, is allowed. The refusal names the directory
  and the chmod that fixes it. Write access granted by an ACL is not detected.
• A scope's exclude_paths must be a list of paths. A single string, an empty
  item, or an item holding a line break or a non-ASCII byte is refused under
  scope-file; before, each was silently dropped, and ZAP could request a URL
  the scope meant to exclude.

What changed:

• Every command speaks through one output system. A terminal gets colour,
  aligned rows and a held-back view of bookkeeping; a pipe or a CI log gets
  every line, plain. -q prints only what needs you (refusals, failures,
  SUSPECT, NOT CHECKED, the verdict); -v prints everything; --no-color,
  --plain and --no-input do what they say. docs/CI.md lists them.
• Refusals are one block: REFUSED, the rule word a CI job can match, and what
  to do. docs/CI.md lists every rule word and the commands that print it.
• argus init asks its questions in place, answered with the arrow keys or by
  typing; piped, it asks nothing. It asks git whether .argus/ is ignored and
  adds the entry when nothing ignores it.
• gate and review report a scanner that is not installed, left an empty or
  misshapen report, or could not read a directory, as NOT CHECKED, never as
  clean.
• dast streams each stage live and summarises every SUSPECT and failed stage,
  even under -q. Its exit is 3 over 1 over 4, whatever order the stages ran.
  The one file it writes, operator.md, is only ever created, in the directory
  it judged and holds for the whole run; each scanner runs in that same
  directory, and a replaced --out stops the run and says so. The exclude list
  is printed in ZAP's run, the one tool that honours it, and one line says the
  other tools do not.
• A run writes only files it creates, never through a link or a second name,
  and never into argus's own inputs.
• Whether evidence is gitignored is asked of the repository the evidence is
  actually in, without running that repository's hooks or fsmonitor, and
  without trusting GIT_DIR and friends from the environment.
• A command printed for you to paste stays one inert command, whatever bytes a
  path holds.

Known, and not in this candidate: terminal control bytes in a path still reach
the screen in plain rows (#121); a handful of other refusal gaps are listed in

This release is not signed. SHA256SUMS catches a corrupt download and proves
nothing about who built the bytes; provenance is the decision docs/adr/0008
leaves open.
-----BEGIN PGP SIGNATURE-----

iJEEABYKADkWIQTKKELDW9dCd1WCColBoKZCVabq8AUCarbUxBsUgAAAAAAEAA5t
YW51MiwyLjUrMS4xMiwwLDMACgkQQaCmQlWm6vDkXgD+MTzoqiTeP1sSIc1jJ5mq
jCwXw4xYDJ7HVyop/WbKvcEBAPhD4xJ5r2+YDnT+v76GGG3mkAIKTiy7P3k216Uo
+pMM
=Jelz
-----END PGP SIGNATURE-----