CI step `format` failed (pipeline 2255, exit 1) -- not on the test, on formatting.
I enumerated the new test by loading packages/mosaic/package.json with python's
json module and writing it back. json.dumps defaults to ensure_ascii=True, so the
em-dash in
"description": "Mosaic agent framework — installation wizard and meta package"
was rewritten as —. Valid JSON, identical when parsed, and rejected by
`pnpm format:check`.
Enumeration is a TEXT edit now: read the file, splice the test into the
test:framework-shell string, write it back. A JSON round-trip rewrites the whole
document to serialiser defaults; the only safe edit to a formatted file is one
that touches the bytes it means to touch.
diff vs main is now exactly one line in each branch.
be-coder-08 found two defects in the regression test I added with the fix. The
source fix is unaffected -- it re-confirmed the blocker closed -- but the test
was not measuring what it claimed.
1. `out=$( ... ) 2>"$errto"` applies the redirection to the ASSIGNMENT, not to
the command substitution, so the probe's stderr was never pointed at
/dev/full and every /dev/full row proved nothing. Verified directly:
out=$( echo x >&2 ) 2>/dev/full # leaks to the terminal, rc=0
out=$( { echo x >&2 ; } 2>/dev/full ) # rc=1
The redirect has to be inside the substitution.
2. `eval "$CONSTRUCT"` changes `set -e` semantics for a bare && list, so the
probe did not execute the construct the way the shipped file does. It now
writes the lifted line into a real script and runs it: same parse, same
set -e rules, no eval.
Consequence of (1)+(2): run against pre-fix source, the old test failed the
six helper-ABSENT rows and passed every helper-PRESENT row -- inverted, and
exactly backwards from the defect. It would have gone green on a broken tree
for the wrong reason.
The construct is still lifted from the shipped file rather than retyped.
Faithful control, matching be-coder-08's prediction exactly:
fixed source: all 12 behavioural rows 0:yes
pre-fix source: 1:no on helper-present + failing-stderr ONLY (3 rows, one per
call site); all 9 other rows 0:yes
Reported-by: be-coder-08
be-coder-08, reviewing #1086, found the diagnostic is not diagnostic-only.
At all three call sites it was written as the last command of an && list:
declare -F explain_tea_user_does_not_exist >/dev/null && explain_tea_user_does_not_exist
and it sits immediately BEFORE the Gitea API fallback. Under `set -e` a failing
diagnostic (stderr closed or full) therefore exits the script and the fallback
never runs -- a diagnostic that suppresses the recovery path it exists to explain.
The asymmetry is what makes it dangerous: the fault only appears when the helper
is PRESENT, so the helper-absent path -- the pre-#1086 behaviour -- keeps working
and reads as a passing control. Measured on /dev/full:
helper present rc=1 fallback NOT reached
helper absent rc=0 fallback reached
Wrapping in `{ ...; } || true` makes the diagnostic status-neutral, which is what
the PR claimed to be in the first place.
test-explain-diagnostic-status-neutral.sh probes all four combinations of
{helper present, absent} x {stderr ok, failing} for each of the three call sites,
and lifts the construct FROM THE SHIPPED FILE rather than restating it -- a probe
that retypes the fixed form passes on a build whose real call sites still carry
the bare && form. Verified RED on the pre-fix tree (16 failures, behavioural half
included) and GREEN here. Enumerated on test:framework-shell.
Reported-by: be-coder-08
- plugins/macp/src/index.ts: use createRequire + dynamic import() for OC SDK
- plugins/macp/src/acp-runtime-types.ts: local ACP runtime type definitions
- plugins/macp/src/macp-runtime.ts: DEFAULT_REPO_ROOT and PI_RUNNER_PATH use
os.homedir() instead of hardcoded /home/user/
- plugins/mosaic-framework/src/index.ts: removed hardcoded SDK import
- No hardcoded /home/ paths remain in any plugin source file
- Plugin works on any machine with openclaw installed globally
Mos