LINBEE-30174 | fix: no prompt nudge on auto-fix or after the skill ran (v1.4.1)

[nivSwisa1]

"turn on auto-fix" tripped the "fix" keyword and nudged a session that was already graded; now auto-fix/autofix is ignored and prompts aren't nudged once the skill ran.

🤖 Generated with Claude Code

✨ PR Description

Purpose: Prevent redundant or false-positive prompt nudges by refining code-task detection and tracking graded repositories.

Main changes:

• Updated keyword detection to ignore "auto-fix" variations using sed-based filtering in trigger script.
• Implemented repo-specific marker logic to suppress nudges after a skill verdict is issued.
• Added test cases for autofix-filtering and repository-scoped status tracking in hooks_test.sh.

Generated by LinearB AI and added by gitStream.
_{AI-generated content may contain inaccuracies. Please verify before using.
💡 Tip: You can customize your AI Description using Guidelines Learn how}

[orca-security-us]

Orca Security Scan Summary

Status	Check	Issues by priority
Passed	Infrastructure as Code	0   0   0   0	View in Orca
Passed	OSS Licenses	0   0   0   0	View in Orca
Passed	SAST	0   0   0   0	View in Orca
Passed	Secrets	0   0   0   0	View in Orca
Passed	Vulnerabilities	0   0   0   0	View in Orca

[gitstream-cm]

✨ PR Review

The shell fix for the false "auto-fix" nudge and skip-after-skill-ran logic looks correct and minimal. However, the PR's ticket reference points to a sub-task about telemetry decision reporting (PreToolUse reporting + session dedup), which has no relation to this nudge-keyword fix, suggesting a ticket/scope mismatch.

1 issues detected:

🎯 Scope - PR implementation does not address the linked ticket's requirements (verdict reporting on PreToolUse, session dedup).

Details: The PR is tagged with LINBEE-30174, whose linked ticket describes reporting the grade on the first PreToolUse after the verdict and deduplicating preliminary/corrected grades via a session key. The actual diff only fixes a UserPromptSubmit nudge false-positive on "auto-fix" and skips nudging after the skill has run — unrelated functionality to the ticket's stated requirements.

File: plugins/agentic-advisor/hooks/agentic-advisor-trigger.sh (126-140)

Generated by LinearB AI and added by gitStream.
_{AI-generated content may contain inaccuracies. Please verify before using.
💡 Tip: You can customize your AI Review using Guidelines Learn how}

[Copilot]

🟡 Changes recommended

The session-wide skill check incorrectly suppresses grading when switching repositories.

1 open finding

• Track grading trigger state separately for each repository · New

What changed in this PR

Prevents false advisor nudges after auto-fix mentions or prior skill execution.

Changes:

• Excludes auto-fix/autofix from keyword detection.
• Suppresses prompts after skill execution.
• Adds regression tests and bumps version to 1.4.1.

File	Description
tests/​agentic-advisor/​hooks_test.sh	Adds prompt-trigger regression tests.
plugins/​agentic-advisor/​hooks/​agentic-advisor-trigger.sh	Updates nudge suppression logic.
plugins/​agentic-advisor/​CHANGELOG.md	Documents the fix.
plugins/​agentic-advisor/​.claude-plugin/​plugin.json	Bumps plugin version.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[gitstream-cm]

✨ PR Review

The fix narrows the "fix" keyword trigger to avoid matching auto-fix variants and adds an early-exit when the skill already ran, writing a done marker so other repo nudges aren't skipped. The core logic looks reasonable; a couple of readability/maintainability points are worth a second look.

2 issues detected:

🧾 Readability - Combining text normalization and keyword matching in one dense pipeline reduces clarity and maintainability. 🛠️

Details: The Tier 2 keyword check now chains a sed substitution and a grep in a single conditional expression, making the matching logic harder to follow and modify (e.g. extending the auto-fix exclusion list or adding more keyword variants).

File: plugins/agentic-advisor/hooks/agentic-advisor-trigger.sh (136-136)

_{🛠️ A suggested code correction is included in the review comments.}

🧹 Maintainability - Relying on an implicit side-effect variable from set_marker to immediately re-write the same marker file is fragile and unclear.

Details: set_marker "$dir" is called and then the code manually truncates/creates "$marker" immediately after. If set_marker already performs an equivalent write as part of its implementation, this extra line is redundant and could mask double-write assumptions as the code evolves; if it doesn't, the dependency on an externally-defined $marker variable set as a side effect of set_marker is implicit and easy to break.

File: plugins/agentic-advisor/hooks/agentic-advisor-trigger.sh (200-200)

Generated by LinearB AI and added by gitStream.
_{AI-generated content may contain inaccuracies. Please verify before using.
💡 Tip: You can customize your AI Review using Guidelines Learn how}

[gitstream-cm]

🧾 Readability - Complex Inline Pipeline: Extract the sed-stripped prompt into an intermediate variable before running the grep check, improving readability and making future keyword additions easier to review.

Suggested change

	elif printf '%s' "$prompt" | sed -E 's/[Aa][Uu][Tt][Oo][- ]?[Ff][Ii][Xx]//g' | grep -qiE '\b(fix|fixes|fixing|fixed|implement|implementing|refactor|refactoring|hotfix|bug ?fix|debug|patch|endpoint|migration)\b'; then
	elif stripped_prompt=$(printf '%s' "$prompt" | sed -E 's/[Aa][Uu][Tt][Oo][- ]?[Ff][Ii][Xx]//g') && printf '%s' "$stripped_prompt" | grep -qiE '\b(fix|fixes|fixing|fixed|implement|implementing|refactor|refactoring|hotfix|bug ?fix|debug|patch|endpoint|migration)\b'; then

Is this review accurate? Use 👍 or 👎 to rate it

If you want to tell us more, use /gs feedback e.g. /gs feedback this review doesn't make sense, I disagree, and it keeps repeating over and over

[nivSwisa1]

Skipping: style only, and the pipeline is a one-liner that matches the surrounding code.

[Copilot]

🟡 Changes recommended

The session-wide skill check can incorrectly mark an ungraded second repository as handled.

1 open finding

• Track skill invocation per repo before creating done marker · New

1 resolved since last review

• Track grading trigger state separately for each repository

🧠 Review effort: Balanced

[Copilot]

🟡 Changes recommended

Repository matching and raw transcript scanning can incorrectly suppress or repeat nudges.

1 open finding

• Resolve verdict identity from remote repository, not checkout directory · New

1 resolved since last review

• Track skill invocation per repo before creating done marker

🧠 Review effort: Balanced

[gitstream-cm]

🚦 Automatic AI Review skipped

This pull request has used all 3 of its automatic AI Reviews.

✨ Comment /gs review to run one now.
_{No arguments needed. It uses the same settings as your automatic reviews: gitStream Settings in Managed Mode, your CM files in Self-Managed.}

[Copilot]

🔵 Needs a closer look

Cross-repository edits can bypass grading, and unstructured transcript matching can falsely mark repositories handled.

0 open findings

1 resolved since last review

• Resolve verdict identity from remote repository, not checkout directory

Previously missed (1)

In code that hasn't changed since last review

Unconditional exit bypasses per-repository grade hold

plugins/​agentic-advisor/​hooks/​agentic-advisor-trigger.sh:208

This exit 0 is reached after any advisor Skill call in the session, even when neither candidate name matched a verdict. As a result, after grading repo A, the first source edit in repo B is allowed instead of hitting the per-repo grade-before-work hold; the new other-repo test checks only the later prompt and misses this bypass. Exit only on a matching verdict and otherwise fall through to the existing .held logic.

This issue also appears on line 205 of the same file.

🧠 Review effort: Balanced

[MishaKav]

LGTM

[Copilot]

🟡 Changes recommended

Raw transcript matching can mistake documentation examples for completed grades and suppress nudges in ungraded repositories.

1 open finding

• Avoid matching verdict examples in non-assistant transcript content · New

🧠 Review effort: Balanced