Repository navigation
docs(unic-archon-dlc): the README config table omits two pr-review keys the Box reads #636
Description
Problem
unic-dlc-pr-review.yaml reads two config keys that the plugin README's config table does not list: pr-review.confidence_threshold, default 60, and pr-review.inline_comments, default true. They are in the Box's header comment, lines 40 and 41 on develop at 6abc1b15. A grep of README.md and commands/setup.md finds neither key. So a Consumer that configures the review Box from the README cannot see either setting or its default.
Found in a review of a second Consumer install, 2026-10-02.
Acceptance criteria
- The README config table has one row for each of the two keys, with its default and what it changes, read from the Box and not from its header comment alone.
- Every config key that any of the four Box YAMLs reads appears in the README config table. Check by reading each Box's config reads, not only the two keys named here. A key the table omits on purpose says why.
- If
commands/setup.mdStep 6 lists config keys, it lists the same set.
Activity
MotionComplex commentedon Oct 5, 2026
A second data point for AC 3, from two Consumers on 0.29.0:
/setupwrote to.archon/unic-dlc.config.yaml6c98622)a6024cca0)project,docs,design,gates,sdlc_needs,build.coverage_threshold,estimations,model_profileSame version, same command, different key sets.
setup.mdStep 6 supports both outcomes: its field list names only the short set, and the same step says to "merge defaults, then what is on disk, then this run's answers". The field list bounds what setup asks. Nothing says which keys it writes. So completing the list (AC 3) still leaves an agent free to write only the keys it asked about.Runtime is unaffected, because a Box falls back to the default. The cost is drift: a later
reconfigurereports the missing keys, and reviewers flag them. DXP-Website added them by hand in PR !5894.Proposed extra acceptance criterion:
One way to get there is to ship the defaults as a YAML file that Step 6 fills in. #330 was folded into #381 on that assumption, but 0.29.0 ships no such file.