Agents open pull requests as WIP and mark them ready for the AI review (v1.37.8) #127

2026-10-06T06:24:25+02:00

claude-code commented

2026-10-06 06:24:25 +02:00

See docs/prs/.

🤖 Generated with Claude Code

See docs/prs/. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

claude-code added 1 commit 2026-10-06 06:24:26 +02:00

PR#33 doc: agents open pull requests as WIP and mark them ready for the AI review

werkator build succeeded after 02:51

Details

61e522fa2e

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

mi added 1 commit 2026-10-06 06:24:40 +02:00

Agents open pull requests as WIP and mark them ready for the AI review (v1.37.8)

ai-review/assessment human review required; a human approval unlocks the merge; review requested from mi

Details

ai-review/gate build succeeded after 00:15

Details

92987ff553

AGENTS.md describes the workflow of the branch protection with ai-review/gate: claude-code opens a WIP pull request,
tools/pr-ready of werkrevisor removes the prefix and schedules the merge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

claude-code marked the pull request as ready for review 2026-10-06 06:24:45 +02:00

claude-code scheduled this pull request to auto merge when all checks succeed 2026-10-06 06:24:45 +02:00

ai-reviewer commented

2026-10-06 06:25:07 +02:00

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

YELLOW risky paths changed: build.gradle.kts (build.gradle.kts)
YELLOW spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md Scenario#127.01 given_when_then_misused (0.64 > yellow_above = 0.6)
YELLOW 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html
hint 2 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37, AGENTS.md:100
hint model recommends merge with caution

Findings

[medium] Scenario mixes actions and outcomes in Then at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37

hint; Jev: real 0.87, severity 1.9 of 3

Scenario#127.01 puts agent behaviour (running tools/pr-ready) and Gitea's merge into the Then block, although they are separate steps. Gitea's merge is an outcome of the checks, not of the agent's action, and the scenario cannot be verified by the stated "this pull request". The Given also does not name the state "ready for review", which the and clause depends on. Split it into two scenarios: the agent opens the PR as WIP, and the agent marks it ready.
[low] Workflow depends on an external repository without a reference at AGENTS.md:100

hint

The text tells agents to run tools/pr-ready "from the werkrevisor repository" without a path or URL. An agent in a fresh session cannot find the tool, so the workflow stays unusable. Add the location or the clone path.

Classifier

Unit	Question	Problem	Confidence	Verdict
docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md Scenario#127.01	given_when_then_misused	0.64	–	YELLOW

4 of 5 units passed all questions (1 PR document(s)).
1 answer ignored because of a confidence below [classifier] min_confidence.

_{Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 20200, calls 1, tokens in 20155 / out 507 (reasoning included in out), duration 12.7 s, cost 0.0454 USD (reported), classifier `jev-1.13.0` calls 5, tokens in 3729, cost 0.0002 USD, finding judge `jev-1.13.0` calls 1, tokens in 1360, cost 0.0001 USD, config: defaults + host.

Reviewed head `92987ff553ec9254f80e2bae79f3e09ebabd76e3` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.}

## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky paths changed: build.gradle.kts (build.gradle.kts) - **YELLOW** spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md Scenario#127.01 given_when_then_misused (0.64 > yellow_above = 0.6) - **YELLOW** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html - **hint** 2 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37, AGENTS.md:100 - **hint** model recommends merge with caution ### Findings 1. **[medium] Scenario mixes actions and outcomes in Then** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37` hint; Jev: real 0.87, severity 1.9 of 3 Scenario#127.01 puts agent behaviour (running `tools/pr-ready`) and Gitea's merge into the `Then` block, although they are separate steps. Gitea's merge is an outcome of the checks, not of the agent's action, and the scenario cannot be verified by the stated "this pull request". The `Given` also does not name the state "ready for review", which the `and` clause depends on. Split it into two scenarios: the agent opens the PR as WIP, and the agent marks it ready. 2. **[low] Workflow depends on an external repository without a reference** at `AGENTS.md:100` hint The text tells agents to run `tools/pr-ready` "from the werkrevisor repository" without a path or URL. An agent in a fresh session cannot find the tool, so the workflow stays unusable. Add the location or the clone path. ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md Scenario#127.01 | given_when_then_misused | 0.64 | – | YELLOW | 4 of 5 units passed all questions (1 PR document(s)). 1 answer ignored because of a confidence below [classifier] min_confidence. Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 20200, calls 1, tokens in 20155 / out 507 (reasoning included in out), duration 12.7 s, cost 0.0454 USD (reported), classifier `jev-1.13.0` calls 5, tokens in 3729, cost 0.0002 USD, finding judge `jev-1.13.0` calls 1, tokens in 1360, cost 0.0001 USD, config: defaults + host. Reviewed head `92987ff553ec9254f80e2bae79f3e09ebabd76e3` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.

ai-reviewer requested review from mi 2026-10-06 06:25:07 +02:00

mi added 1 commit 2026-10-06 06:26:19 +02:00

PR-doc open questions in the format of hsadmin-ng: optional, TODO.<topic> defers, other unchecked items block

werkator build succeeded after 02:44

Details

ai-review/assessment AI findings; a developer approval after checking them unlocks the merge; a human review request is pending

Details

ai-review/gate build succeeded after 00:17

Details

dd8bf2ac42

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ai-reviewer commented

2026-10-06 06:26:44 +02:00

🟠 werkrevisor: ORANGE / FINDINGS

Change kind: mixed.

Reasons

ORANGE spec: docs/prs/TEAMPLATE.md Scenario#236.02 hard_to_read (0.94 > red_above = 0.75)
YELLOW risky paths changed: build.gradle.kts (build.gradle.kts)
YELLOW spec: docs/prs/TEAMPLATE.md Scenario#236.01 vague_then (0.63 > yellow_above = 0.6)
YELLOW spec: docs/prs/TEAMPLATE.md Scenario#236.02 vague_then (0.61 > yellow_above = 0.6)
YELLOW 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html
hint 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:7-10, AGENTS.md:100
hint model recommends merge with caution

Findings

[medium] Scenario names an unrelated-sounding test at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37

hint; Jev: real 0.83, severity 1.9 of 3

Scenario#127.01 is verified only by "this pull request, which follows the workflow". That is not a link to a test or checkable artifact, and the scenario's Then mixes three outcomes (WIP prefix, tools/pr-ready, Gitea merge). The Gitea merge outcome depends on external config and cannot be verified from this repo. Split the scenario or state the check explicitly.
[low] Wrong file name and PR claims inconsistent at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:7-10

hint

The Additional Changes section refers to docs/prs/TEAMPLATE.md, which matches the file in the diff, so this is consistent. However, "Since 2026-10-06 the branch protection ... requires" states a host-side fact that the repo cannot verify, and "it never approves a pull request that its own bot opened" is not reflected in AGENTS.md, which tells agents to open PRs as claude-code and says ai-reviewer approves. This leaves the reader unsure why a green PR from claude-code is approved by the bot. Clarify that claude-code differs from the review bot.
[low] Hard invariant text depends on an external repository at AGENTS.md:100

hint

The new section instructs running tools/pr-ready from the werkrevisor repository, with no path or install location. An agent without that checkout cannot follow the workflow. Name where it lives or how to obtain it.

Classifier

Unit	Question	Problem	Confidence	Verdict
docs/prs/TEAMPLATE.md Scenario#236.01	vague_then	0.63	–	YELLOW
docs/prs/TEAMPLATE.md Scenario#236.02	hard_to_read	0.94	0.81	ORANGE
docs/prs/TEAMPLATE.md Scenario#236.02	vague_then	0.61	–	YELLOW

16 of 18 units passed all questions (2 PR document(s)).
5 answers ignored because of a confidence below [classifier] min_confidence.

_{Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21387, calls 1, tokens in 21370 / out 767 (reasoning included in out), duration 14.2 s, cost 0.0504 USD (reported), classifier `jev-1.13.0` calls 18, tokens in 12326, cost 0.0005 USD, finding judge `jev-1.13.0` calls 1, tokens in 1446, cost 0.0001 USD, config: defaults + host.

Reviewed head `dd8bf2ac424ee8e849bd6106cd8cbf9847e3dad7` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.}

## 🟠 werkrevisor: ORANGE / FINDINGS Change kind: `mixed`. ### Reasons - **ORANGE** spec: docs/prs/TEAMPLATE.md Scenario#236.02 hard_to_read (0.94 > red_above = 0.75) - **YELLOW** risky paths changed: build.gradle.kts (build.gradle.kts) - **YELLOW** spec: docs/prs/TEAMPLATE.md Scenario#236.01 vague_then (0.63 > yellow_above = 0.6) - **YELLOW** spec: docs/prs/TEAMPLATE.md Scenario#236.02 vague_then (0.61 > yellow_above = 0.6) - **YELLOW** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html - **hint** 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:7-10, AGENTS.md:100 - **hint** model recommends merge with caution ### Findings 1. **[medium] Scenario names an unrelated-sounding test** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:25-37` hint; Jev: real 0.83, severity 1.9 of 3 Scenario#127.01 is verified only by "this pull request, which follows the workflow". That is not a link to a test or checkable artifact, and the scenario's `Then` mixes three outcomes (WIP prefix, `tools/pr-ready`, Gitea merge). The Gitea merge outcome depends on external config and cannot be verified from this repo. Split the scenario or state the check explicitly. 2. **[low] Wrong file name and PR claims inconsistent** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:7-10` hint The Additional Changes section refers to `docs/prs/TEAMPLATE.md`, which matches the file in the diff, so this is consistent. However, "Since 2026-10-06 the branch protection ... requires" states a host-side fact that the repo cannot verify, and "it never approves a pull request that its own bot opened" is not reflected in AGENTS.md, which tells agents to open PRs as `claude-code` and says `ai-reviewer` approves. This leaves the reader unsure why a green PR from `claude-code` is approved by the bot. Clarify that `claude-code` differs from the review bot. 3. **[low] Hard invariant text depends on an external repository** at `AGENTS.md:100` hint The new section instructs running `tools/pr-ready` from the werkrevisor repository, with no path or install location. An agent without that checkout cannot follow the workflow. Name where it lives or how to obtain it. ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/TEAMPLATE.md Scenario#236.01 | vague_then | 0.63 | – | YELLOW | | docs/prs/TEAMPLATE.md Scenario#236.02 | hard_to_read | 0.94 | 0.81 | ORANGE | | docs/prs/TEAMPLATE.md Scenario#236.02 | vague_then | 0.61 | – | YELLOW | 16 of 18 units passed all questions (2 PR document(s)). 5 answers ignored because of a confidence below [classifier] min_confidence. Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21387, calls 1, tokens in 21370 / out 767 (reasoning included in out), duration 14.2 s, cost 0.0504 USD (reported), classifier `jev-1.13.0` calls 18, tokens in 12326, cost 0.0005 USD, finding judge `jev-1.13.0` calls 1, tokens in 1446, cost 0.0001 USD, config: defaults + host. Reviewed head `dd8bf2ac424ee8e849bd6106cd8cbf9847e3dad7` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.

mi added 1 commit 2026-10-06 06:28:37 +02:00

review: PR#127 scenario split into opening as WIP and marking as ready, tools/pr-ready located, claude-code told apart from the review bot

werkator build succeeded after 02:16

Details

ai-review/assessment human review required; a human approval unlocks the merge; a human review request is pending

Details

ai-review/gate build succeeded after 00:16

Details

cb8c9e1fb0

Resolves the findings and the YELLOW classifier reason of the werkrevisor review.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ai-reviewer commented

2026-10-06 06:28:54 +02:00

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

YELLOW risky paths changed: build.gradle.kts (build.gradle.kts)
YELLOW spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' hard_to_understand (0.51 > yellow_above = 0.5)
YELLOW 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html
hint 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:51-53, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-61, AGENTS.md:98-101
hint model recommends merge with caution

Findings

[medium] Scenario verified by a test outside this repository at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:51-53

hint; Jev: real 0.80, severity 2.0 of 3

Scenario#127.02 is verified only by a test in werkrevisor, which is a different repository and can change or vanish independently of this PR. The PR doc points at tools/pr-ready, whose behavior this PR does not implement. The link is to main of another repo, so it is neither pinned nor guaranteed to exist at merge time. Name the werkrevisor PR or commit that adds the test, or state explicitly that it is an external verification.
[low] Unclear solution section and filler at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-61

hint

The Solution section is thin and partly filler. "Version 1.37.8, because every PR needs one." narrates a rule and adds nothing. The sentence about fixing YELLOW reasons ("e.g. a scenario that is hard to read") is vague about which reasons count. The section does not mention that the workflow also lives in the pr-doc skill and TEAMPLATE, which sit only under "Additional Changes".
[low] Environment-specific details in shared instructions at AGENTS.md:98-101

hint

The new section hardcodes a local path (~/Projekte/werkrevisor) and account names such as mi and claude-code. AGENTS.md is described as shared and tool-agnostic, so the path will be wrong for other agents or machines. Refer to the repository by URL only, or say where the clone location is configured.

Classifier

Unit	Question	Problem	Confidence	Verdict
docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution'	hard_to_understand	0.51	0.48	YELLOW

7 of 8 units passed all questions (1 PR document(s)).

_{Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21328, calls 1, tokens in 21362 / out 746 (reasoning included in out), duration 13.2 s, cost 0.0502 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 5970, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1297, cost 0.0001 USD, config: defaults + host.

Reviewed head `cb8c9e1fb0a96ad42ced50f11acf009b95165289` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.}

## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky paths changed: build.gradle.kts (build.gradle.kts) - **YELLOW** spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' hard_to_understand (0.51 > yellow_above = 0.5) - **YELLOW** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html - **hint** 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:51-53, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-61, AGENTS.md:98-101 - **hint** model recommends merge with caution ### Findings 1. **[medium] Scenario verified by a test outside this repository** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:51-53` hint; Jev: real 0.80, severity 2.0 of 3 Scenario#127.02 is verified only by a test in werkrevisor, which is a different repository and can change or vanish independently of this PR. The PR doc points at `tools/pr-ready`, whose behavior this PR does not implement. The link is to `main` of another repo, so it is neither pinned nor guaranteed to exist at merge time. Name the werkrevisor PR or commit that adds the test, or state explicitly that it is an external verification. 2. **[low] Unclear solution section and filler** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-61` hint The Solution section is thin and partly filler. "Version 1.37.8, because every PR needs one." narrates a rule and adds nothing. The sentence about fixing YELLOW reasons ("e.g. a scenario that is hard to read") is vague about which reasons count. The section does not mention that the workflow also lives in the pr-doc skill and TEAMPLATE, which sit only under "Additional Changes". 3. **[low] Environment-specific details in shared instructions** at `AGENTS.md:98-101` hint The new section hardcodes a local path (`~/Projekte/werkrevisor`) and account names such as `mi` and `claude-code`. AGENTS.md is described as shared and tool-agnostic, so the path will be wrong for other agents or machines. Refer to the repository by URL only, or say where the clone location is configured. ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' | hard_to_understand | 0.51 | 0.48 | YELLOW | 7 of 8 units passed all questions (1 PR document(s)). Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21328, calls 1, tokens in 21362 / out 746 (reasoning included in out), duration 13.2 s, cost 0.0502 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 5970, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1297, cost 0.0001 USD, config: defaults + host. Reviewed head `cb8c9e1fb0a96ad42ced50f11acf009b95165289` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.

mi added 1 commit 2026-10-06 07:25:05 +02:00

PR#127: the change of the title from WIP starts the review, the bot schedules the merge; no tools/pr-ready

ai-review/assessment human review required; a human approval unlocks the merge; a human review request is pending

Details

ai-review/gate build succeeded after 00:16

Details

werkator build succeeded after 02:25

Details

cc1fcb85b9

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ai-reviewer commented

2026-10-06 07:25:25 +02:00

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

YELLOW risky paths changed: build.gradle.kts (build.gradle.kts)
YELLOW spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' hard_to_understand (0.50 > yellow_above = 0.5)
YELLOW 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html
hint 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:42-49, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-67, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:38-40
hint model recommends merge with caution

Findings

[medium] Contradiction on merge after YELLOW/ORANGE at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:42-49

hint; Jev: real 0.84, severity 2.3 of 3

Scenario#127.02 says the bot schedules the merge after a GREEN, YELLOW or ORANGE review. AGENTS.md says Gitea merges GREEN at once but YELLOW/ORANGE only when mi approves. The scenario is therefore wrong about the unattended merge ("merges without a click"). It should state that YELLOW/ORANGE wait for an approval by mi.
[low] Wrong filename in "Additional Changes" at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-67

hint

The document refers to docs/prs/TEAMPLATE.md, which is the actual (misspelled) filename in the diff, but readers may take it for a typo. Also the "The Solution" section lists "Version 1.37.8, because every PR needs one." as filler, and it omits that the pr-doc skill and AGENTS.md are changed in this PR. The section is hard to follow (classifier hint).
[low] Scenario#127.01 verification is not a test at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:38-40

hint

"Verified by" points to the instructions themselves and says this PR follows them, so nothing verifies the scenario. A reference to the title of this PR, or an explicit note that the scenario is process-only, would be clearer.

Classifier

Unit	Question	Problem	Confidence	Verdict
docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution'	hard_to_understand	0.50	0.48	YELLOW

7 of 8 units passed all questions (1 PR document(s)).
2 answers ignored because of a confidence below [classifier] min_confidence.

_{Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21542, calls 1, tokens in 21544 / out 743 (reasoning included in out), duration 14.2 s, cost 0.0505 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 5972, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1410, cost 0.0001 USD, config: defaults + host.

Reviewed head `cc1fcb85b9b5014476ff20d48725e291d89fc3b4` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.}

## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky paths changed: build.gradle.kts (build.gradle.kts) - **YELLOW** spec: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' hard_to_understand (0.50 > yellow_above = 0.5) - **YELLOW** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html - **hint** 3 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:42-49, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-67, docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:38-40 - **hint** model recommends merge with caution ### Findings 1. **[medium] Contradiction on merge after YELLOW/ORANGE** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:42-49` hint; Jev: real 0.84, severity 2.3 of 3 Scenario#127.02 says the bot schedules the merge after a GREEN, YELLOW or ORANGE review. AGENTS.md says Gitea merges GREEN at once but YELLOW/ORANGE only when `mi` approves. The scenario is therefore wrong about the unattended merge ("merges without a click"). It should state that YELLOW/ORANGE wait for an approval by `mi`. 2. **[low] Wrong filename in "Additional Changes"** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:55-67` hint The document refers to `docs/prs/TEAMPLATE.md`, which is the actual (misspelled) filename in the diff, but readers may take it for a typo. Also the "The Solution" section lists "Version 1.37.8, because every PR needs one." as filler, and it omits that the `pr-doc` skill and `AGENTS.md` are changed in this PR. The section is hard to follow (classifier hint). 3. **[low] Scenario#127.01 verification is not a test** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:38-40` hint "Verified by" points to the instructions themselves and says this PR follows them, so nothing verifies the scenario. A reference to the title of this PR, or an explicit note that the scenario is process-only, would be clearer. ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md section 'The Solution' | hard_to_understand | 0.50 | 0.48 | YELLOW | 7 of 8 units passed all questions (1 PR document(s)). 2 answers ignored because of a confidence below [classifier] min_confidence. Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21542, calls 1, tokens in 21544 / out 743 (reasoning included in out), duration 14.2 s, cost 0.0505 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 5972, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1410, cost 0.0001 USD, config: defaults + host. Reviewed head `cc1fcb85b9b5014476ff20d48725e291d89fc3b4` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.15.0.

mi added 1 commit 2026-10-06 07:39:45 +02:00

review: PR#127 doc names what changes, says that YELLOW and ORANGE wait for mi, and marks the process-only scenario

ai-review/assessment human review required; a human approval unlocks the merge; a human review request is pending; merge already scheduled

Details

ai-review/gate build succeeded after 00:17

Details

werkator build succeeded after 02:54

Details

5191e972bc

Resolves the YELLOW classifier reason and the three hints of the werkrevisor review.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

ai-reviewer commented

2026-10-06 07:39:59 +02:00

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

YELLOW risky paths changed: build.gradle.kts (build.gradle.kts)
YELLOW 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html
hint 2 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:58-65, AGENTS.md:102
hint model recommends merge with caution

Findings

[medium] PR doc and template path inconsistent at docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:58-65

hint; Jev: real 0.67, severity 1.5 of 3

The PR doc says the template docs/prs/TEAMPLATE.md and the pr-doc skill "take over" the open-questions format, and lists this under both "The Solution" and "Additional Changes", which is redundant. It also says "see below" while the explanation is only in the later section. Scenario#127.02 claims the test test_flow.ReadinessAndAutomergeTest verifies the behaviour, but that test lives in another repository and covers werkrevisor, not the AGENTS.md rule. Link and claim cannot be checked from this PR.
[low] Fix-everything rule is vague at AGENTS.md:102

hint

The sentence "every YELLOW reason that you can resolve, e.g. a scenario that is hard to read; leave only structural ones, such as a dangerous path, to mi" gives agents no criterion for "can resolve". Different agents will draw the line differently, so the rule is hard to apply consistently.

Classifier

8 of 8 units passed all questions (1 PR document(s)).
1 answer ignored because of a confidence below [classifier] min_confidence.

_{Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21493, calls 1, tokens in 21494 / out 545 (reasoning included in out), duration 15.0 s, cost 0.0484 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 6021, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1279, cost 0.0001 USD, config: defaults + host.

Reviewed head `5191e972bccd01c136dc073353449b7b34c6637d` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.16.1.}

## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky paths changed: build.gradle.kts (build.gradle.kts) - **YELLOW** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (each larger than [review] max_full_file_chars = 40000): src/main/resources/templates/releases.html - **hint** 2 finding(s) of low severity or rated minor by Jev: docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:58-65, AGENTS.md:102 - **hint** model recommends merge with caution ### Findings 1. **[medium] PR doc and template path inconsistent** at `docs/prs/2026-10-06-PR#127-agents-pr-workflow-with-ai-review.md:58-65` hint; Jev: real 0.67, severity 1.5 of 3 The PR doc says the template `docs/prs/TEAMPLATE.md` and the pr-doc skill "take over" the open-questions format, and lists this under both "The Solution" and "Additional Changes", which is redundant. It also says "see below" while the explanation is only in the later section. Scenario#127.02 claims the test `test_flow.ReadinessAndAutomergeTest` verifies the behaviour, but that test lives in another repository and covers werkrevisor, not the `AGENTS.md` rule. Link and claim cannot be checked from this PR. 2. **[low] Fix-everything rule is vague** at `AGENTS.md:102` hint The sentence "every YELLOW reason that you can resolve, e.g. a scenario that is hard to read; leave only structural ones, such as a dangerous path, to `mi`" gives agents no criterion for "can resolve". Different agents will draw the line differently, so the rule is hard to apply consistently. ### Classifier 8 of 8 units passed all questions (1 PR document(s)). 1 answer ignored because of a confidence below [classifier] min_confidence. Usage: model `openrouter/anthropic/claude-sonnet-5.5`, LLM done, tokens counted before the call 21493, calls 1, tokens in 21494 / out 545 (reasoning included in out), duration 15.0 s, cost 0.0484 USD (reported), classifier `jev-1.13.0` calls 8, tokens in 6021, cost 0.0003 USD, finding judge `jev-1.13.0` calls 1, tokens in 1279, cost 0.0001 USD, config: defaults + host. Reviewed head `5191e972bccd01c136dc073353449b7b34c6637d` against base `87d40f34dadee4b7d3a7febca854d27ed187fa17`, werkrevisor 0.16.1.

mi merged commit 0e9f13edbf into main

2026-10-06 07:58:29 +02:00

mi deleted branch claude-code/agents-pr-workflow-with-ai-review

2026-10-06 07:58:29 +02:00

Sign in to join this conversation.

3 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: mi/werkator#127