fix(ploeg): show verification from the worker's record, not agent prose

ryangr0 commented

2026-10-03 00:19:46 +00:00

Owner

The worker reported its verification only as prose: a
"[Ploeg verification passed|failed|incomplete]" marker appended to the
agent-written summary and a "### Ploeg verification" section appended to
the findings. The usage report took the first marker in the summary and
the first commit hash in the findings, so an agent whose own summary said
"[Ploeg verification passed]" made a failed check render as passed, on
whatever commit the agent named.

The worker now also sends a structured verification record on the
OutcomeReport: the result, the full commit, the dirty-tree flag, why
checks stopped, start and finish times, and each check with its result,
exit status and times. resolveOutcome discards any verification an
adapter or agent reported, so only the worker's own run sets it. ploegd
validates the record, stores it in agent_runs.verification (migration
0034) for writing Runs only, and RoundReports returns it. parseEvidence
uses the record whenever the last writing Run has one; summary and
findings text cannot change it. Dirty, failed, incomplete and unknown
render distinctly.

Runs reported by an older worker have no record and still render from
the prose, now taking the last marker and the last section, which are
the ones the worker appended. The field is optional on the wire, so
older payloads decode unchanged and an older ploegd ignores it. The
outcome report schema and contract docs describe the new field.

Adds an optional verification record to the outcome contract (schema and README updated) and migration 0034 (agent_runs.verification JSONB, writers only). Older payloads decode unchanged and fall back to the prose, now reading the last marker the worker appended. Not yet exposed in Vloer or the Run APIs (follow-up). A malformed record makes ploegd reject the outcome with 400, like other invalid fields.

Verified with mise run verify on the pinned toolchain (all gates passed).

Ticket: https://vikunja.webgrip.dev/tasks/1733

🤖 Generated with Claude Code

The worker reported its verification only as prose: a "[Ploeg verification passed|failed|incomplete]" marker appended to the agent-written summary and a "### Ploeg verification" section appended to the findings. The usage report took the first marker in the summary and the first commit hash in the findings, so an agent whose own summary said "[Ploeg verification passed]" made a failed check render as passed, on whatever commit the agent named. The worker now also sends a structured verification record on the OutcomeReport: the result, the full commit, the dirty-tree flag, why checks stopped, start and finish times, and each check with its result, exit status and times. resolveOutcome discards any verification an adapter or agent reported, so only the worker's own run sets it. ploegd validates the record, stores it in agent_runs.verification (migration 0034) for writing Runs only, and RoundReports returns it. parseEvidence uses the record whenever the last writing Run has one; summary and findings text cannot change it. Dirty, failed, incomplete and unknown render distinctly. Runs reported by an older worker have no record and still render from the prose, now taking the last marker and the last section, which are the ones the worker appended. The field is optional on the wire, so older payloads decode unchanged and an older ploegd ignores it. The outcome report schema and contract docs describe the new field. Adds an optional `verification` record to the outcome contract (schema and README updated) and migration 0034 (agent_runs.verification JSONB, writers only). Older payloads decode unchanged and fall back to the prose, now reading the last marker the worker appended. Not yet exposed in Vloer or the Run APIs (follow-up). A malformed record makes ploegd reject the outcome with 400, like other invalid fields. Verified with `mise run verify` on the pinned toolchain (all gates passed). Ticket: https://vikunja.webgrip.dev/tasks/1733 🤖 Generated with [Claude Code](https://claude.com/claude-code)

ryangr0 added 1 commit

2026-10-03 00:19:46 +00:00

fix(ploeg): show verification from the worker's record, not agent prose

[Workflow] On Pull Request / checks (pull_request) Has been cancelled

Details

[Workflow] On Pull Request / warnings (pull_request) Has been cancelled

Details

[Workflow] On Pull Request / release-policy (pull_request) Has been cancelled

Details

7bfd8a34fb

The worker reported its verification only as prose: a
"[Ploeg verification passed|failed|incomplete]" marker appended to the
agent-written summary and a "### Ploeg verification" section appended to
the findings. The usage report took the first marker in the summary and
the first commit hash in the findings, so an agent whose own summary said
"[Ploeg verification passed]" made a failed check render as passed, on
whatever commit the agent named.

The worker now also sends a structured verification record on the
OutcomeReport: the result, the full commit, the dirty-tree flag, why
checks stopped, start and finish times, and each check with its result,
exit status and times. resolveOutcome discards any verification an
adapter or agent reported, so only the worker's own run sets it. ploegd
validates the record, stores it in agent_runs.verification (migration
0034) for writing Runs only, and RoundReports returns it. parseEvidence
uses the record whenever the last writing Run has one; summary and
findings text cannot change it. Dirty, failed, incomplete and unknown
render distinctly.

Runs reported by an older worker have no record and still render from
the prose, now taking the last marker and the last section, which are
the ones the worker appended. The field is optional on the wire, so
older payloads decode unchanged and an older ploegd ignores it. The
outcome report schema and contract docs describe the new field.

VIK-1733

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ryangr0 merged commit 855022c634 into development

2026-10-03 00:21:49 +00:00

ryangr0 referenced this pull request from a commit

2026-10-03 00:21:50 +00:00

Merge pull request 'fix(ploeg): show verification from the worker's record, not agent prose' (#152) from ryangr0/ploeg-structured-verification into development

ryangr0 referenced this pull request

2026-10-03 11:52:30 +00:00

docs(ploeg): propose ADR-0059 delivery facts come from the forge #173

ryangr0 referenced this pull request from a commit

2026-10-03 11:54:42 +00:00

docs(ploeg): propose ADR-0059 delivery facts come from the forge

ryangr0 referenced this pull request