@mrciphersmith/keryx 0.3.124 → 0.3.125
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +4 -0
- package/dist/cli.js +494 -43
- package/dist/core.js +59 -10
- package/package.json +1 -1
- package/src/gdskills/bundled/skills/orchestration/flow-orchestrator/SKILL.md +1 -1
- package/src/gdskills/bundled/skills/orchestration/job-orchestrator/SKILL.md +5 -5
- package/src/gdskills/bundled/skills/review/review-orchestrator/SKILL.detail.md +8 -8
- package/src/gdskills/bundled/skills/review/review-orchestrator/SKILL.md +18 -19
- package/src/gdskills/bundled/skills/review/review-premise/SKILL.md +2 -2
- package/src/gdskills/contracts/review-finding.schema.json +14 -1
package/README.md
CHANGED
|
@@ -1176,6 +1176,10 @@ Every managed review start, attach or ingest (and any `dispatch-check` or `blast
|
|
|
1176
1176
|
|
|
1177
1177
|
`keryx review complete` on a pull-request package (one that targets a PR, or whose `--ref` is a PR URL under any `--target`) refuses a merged PR or a moved head until the first line of `report.md` names the reviewed SHA and says so, and refuses a report with no `Evidence level:` line (`static only — did not run: <what>` or `ran — <what>`; a package ingested before the field existed only warns). `keryx review ingest` reads each finding's quote at its `path` at the reviewed SHA and refuses one that does not resolve there, adds a `citation not found` line to a rebuttal that cites a rule or document missing from the tree (the finding is never deleted), and a finding can close as `narrowed` only with a "what remains" text. See [review with a record](docs/docs/guides/review-with-a-record.md).
|
|
1178
1178
|
|
|
1179
|
+
### Operator decisions in a review: a card, or a host question
|
|
1180
|
+
|
|
1181
|
+
Four review acts are the operator's: `keryx review confirm-plan`, `waive-mutation`, `override-reviewers` and `decide-premise` (a held finding's premise). In `keryx shell` the agent calls a card and you answer it in the shell or in the Telegram topic; a card press is verified by keryx. In Claude Code, Codex or another host there is no card: the agent asks you with the host's own question tool (or a Telegram poll through Helyx) and records your answer with `--answered-by <source>:<ref> --answer-text "<your words, verbatim>" --question-text "<the question as asked>"`, with no terminal. That record carries `confirmedVia: "host-question"` and is weaker than a card: keryx does not verify the answer. Free text is never a decision, an unattended run refuses the flag, and `keryx review complete` with its waivers stays on the card or terminal path. See [limitations](docs/docs/limitations.md#a-host-question-is-a-weaker-proof-than-a-card) and the [CLI reference](docs/docs/cli-reference.md#review).
|
|
1182
|
+
|
|
1179
1183
|
### Review bot on pull requests
|
|
1180
1184
|
|
|
1181
1185
|
A GitHub Action reviews each same-repository pull request, has a second model
|