Filter issue_comment events containing slash commands from review dispatch

[fullsend-ai-retro]

What happened

On PR #6906, the PR author posted /review at 09:32Z (a slash command intended for the qodo-code-review bot). This triggered fullsend.yaml workflow run 26025258675, which dispatched a review. The review agent re-analyzed the same code it had already reviewed 25 minutes earlier, producing identical findings. A "fullsend review is working on this" notification was posted, adding noise.

This is distinct from the event-type filtering tracked in #893, which focuses on pull_request_review events. This case involves issue_comment events where the comment body is a slash command (e.g., /review, /retest, /ask) meant for other bots.

What could go better

The dispatch layer should recognize that issue comments starting with / are typically slash commands for other tools (qodo, prow, etc.) and not signals that new review is needed. On this PR, 1 of the 4 review dispatches was triggered this way. Looking at #893, the same PR #816 that motivated that issue likely also had slash-command-triggered dispatches.

Confidence: High that this is a real waste pattern. Slash commands from humans in PR comments are never meaningful triggers for fullsend review. The only issue_comment events that should trigger review are ones where the comment explicitly requests fullsend action (e.g., /fullsend review).

Proposed change

In the fullsend.yaml shim workflow's dispatch-review job, add a condition to skip dispatch when the trigger is issue_comment and the comment body starts with / (unless it starts with /fullsend). This could be implemented as:

if: |
  github.event_name != 'issue_comment' ||
  !startsWith(github.event.comment.body, '/') ||
  startsWith(github.event.comment.body, '/fullsend')

This complements #893 (which filters pull_request_review events) and #981 (which adds concurrency groups). Together these three changes would have reduced the 4 review dispatches on PR #6906 to 1-2.

Alternatively, this filter could be added to the dispatch layer itself (#766) if that architectural change lands first.

Validation criteria

1. On a test PR, posting /review, /retest, /lgtm, or other non-fullsend slash commands should NOT trigger a fullsend review dispatch.
2. Posting /fullsend review should still trigger a dispatch.
3. Measure across 20 PRs in konflux-ci/konflux-ci: the number of review dispatches triggered by issue_comment events with slash-command bodies should drop to zero (excluding /fullsend commands).

---

Generated by retro agent from konflux-ci/konflux-ci#6906

[fullsend-ai-triage]

Triage Summary

Severity: Medium · Category: Feature

Problem

Slash commands posted in PR comments (e.g., /review, /retest, /lgtm) intended for other bots (qodo, prow) are triggering fullsend review dispatches via issue_comment events. This produces redundant reviews and notification noise.

Root Cause

The dispatch-review job in fullsend.yaml does not filter issue_comment events by comment body content, so any comment triggers a dispatch.

Recommended Fix

Add a conditional to the dispatch-review job:

if: |
  github.event_name != 'issue_comment' ||
  !startsWith(github.event.comment.body, '/') ||
  startsWith(github.event.comment.body, '/fullsend')

This complements #893 (filtering pull_request_review events) and #981 (concurrency groups). If the dispatch-layer refactor in #766 lands first, this filter could be applied there instead.

Proposed Test Case

1. issue_comment with body '/review'        → no dispatch (filtered)
2. issue_comment with body '/fullsend review' → dispatch (allowed)
3. issue_comment with body 'LGTM'           → dispatch (allowed)
4. pull_request event                        → dispatch (unaffected)
5. Edge case: comment body is just '/'       → no dispatch (filtered)

Relationship to Other Issues

• Reduce redundant review dispatches triggered by non-code events #893 — Filters pull_request_review events (different event type, complementary)
• Add concurrency group to dispatch-review job in fullsend.yaml #981 — Adds concurrency groups (orthogonal dedup mechanism)
• feat: dispatch-layer dedup and per-issue concurrency guard to prevent duplicate agent runs #766 — Dispatch-layer dedup (potential alternative location for this filter)

---

Labels: This issue affects the dispatch layer (area/dispatch) and the review agent pipeline (agent/review). It is a maintenance enhancement (type/chore) to reduce noise, with medium priority since it causes redundant work but not incorrect behavior. Tagged agent/retro as the source.

[fullsend-ai-prioritize]

RICE Priority Score: 1.35

Score breakdown

Dimension	Score	Reasoning
Reach	1.5	This affects any repo using fullsend alongside other bots that use slash commands (qodo, prow, etc.). The issue was surfaced from konflux-ci but the pattern applies across multiple orgs and customers using similar tooling setups, warranting a 1.5 score.
Impact	0.5	The issue causes redundant review dispatches and notification noise but does not produce incorrect behavior. Workarounds exist (ignore the noise), and the redundant reviews are identical to prior ones. This is a low-impact annoyance rather than a blocking problem.
Confidence	0.9	Very high confidence — the issue is extremely well-described with a specific real-world example (PR #6906), a concrete proposed fix (a YAML conditional), clear validation criteria, and relationships to related issues (#893, #981, #766) are documented.
Effort	0.5	Simple change — adding a conditional expression to the dispatch-review job in fullsend.yaml. The proposed implementation is a 3-line YAML condition. Testing requires verifying slash commands are filtered while /fullsend commands still work.

Formula: (1.5 x 0.5 x 0.9) / 0.5 = 1.35