Issue Triage

TizenFX Issue Triage Pipeline

Overview

For open issues in samsung/TizenFX that have not been claimed by a human owner and are not already in the AI execution queue (ai-task label):

1. Post a first-pass technical response in English summarising the issue and offering an initial analysis or pointer.
2. Identify the most likely code area, look up .github/CODEOWNERS, and mention the matching reviewer(s) so they can self-assign.

This skill never:

• modifies code,
• creates PRs or branches,
• assigns issues (gh issue edit --add-assignee),
• adds or removes labels,
• closes issues.

It only comments. All decisions stay with the humans involved.

All comments are written in English and prefixed with 🤖 [AI Triage] so future runs can skip already-triaged issues.

Repository

• Repo: samsung/TizenFX (GitHub)
• CLI: gh CLI (authenticated)
• CODEOWNERS path: .github/CODEOWNERS in the repo working tree

Constants

MAX_ISSUES_PER_RUN   = 5
STALE_THRESHOLD_DAYS = 90        # Issues with updatedAt older than this are skipped
DEFAULT_FALLBACK     = @JoonghyunCho   # When CODEOWNERS lookup fails
COMMENT_PREFIX       = 🤖 [AI Triage]

Pipeline Flow

⓪ Discover triage candidates
  → ① Apply skip filters
    → ② For each survivor (max 5):
      a. Fetch issue body + comments
      b. Identify likely code area
      c. CODEOWNERS lookup
      d. Draft first-pass technical response
      e. Post comment (mention only — no assign, no label)

Stage ⓪: Discover Open Triage Candidates

gh issue list --repo samsung/TizenFX --state open \
  --json number,title,createdAt,updatedAt,assignees,labels,author \
  --jq '[.[]
         | select((.labels // []) | map(.name) | index("ai-task") | not)
         | select((.assignees // []) | length == 0)]
        | sort_by(.updatedAt) | reverse'

• Drops issues with ai-task label (handled by refactor-execute)
• Drops issues with any assignee (someone is already on it)
• Sorted by updatedAt descending (most recently active first)

Stage ①: Apply Skip Filters

Filter the candidate list down further. In order, drop an issue when any of these holds:

1. Already triaged by AI — issue has at least one comment whose body starts with 🤖 [AI Triage].

   gh issue view {N} --repo samsung/TizenFX --comments \
     --json comments --jq '.comments[] | select(.body | startswith("🤖 [AI Triage]"))' \
     | head -1

   Non-empty result → skip.

2. Stale — updatedAt is older than STALE_THRESHOLD_DAYS (90 days).

   • Compute: (now - updatedAt) in days. If > 90, skip.
   • Reason: Re-engaging a long-dormant thread creates noise; re-activation is a human decision.

3. Author is a bot — author.login ends with [bot] or matches a known bot list (dependabot, renovate, github-actions).

   • Skip; bots typically don't need triage answers.

After filters, take up to MAX_ISSUES_PER_RUN (5) issues, oldest-updatedAt-first among the survivors so we drain the older ones first.

Stage ②a: Fetch Issue Detail

For each selected issue:

gh issue view {N} --repo samsung/TizenFX --comments \
  --json title,body,comments,labels,author

Read all comments (not only the body) so we don't repeat what someone has already answered.

Stage ②b: Identify Likely Code Area

Goal: produce a list of candidate repo paths for CODEOWNERS lookup.

Heuristics (apply all that match):

1. Explicit code references in body — file paths like src/Tizen.Sensor/..., Tizen.NUI.Layout, fully-qualified class names like BluetoothErrorFactory.
2. Title keywords — namespace fragments (NUI, Sensor, Bluetooth, Application) map to src/Tizen.NUI/, src/Tizen.Sensor/, etc.
3. Code blocks — using Tizen.Foo; reveals the namespace, which maps to a src/Tizen.Foo/ directory (verify by ls src/Tizen.Foo* if uncertain).

If no path can be inferred, emit a single placeholder candidate of (unknown) and proceed to Stage ②c with the fallback owner.

Cap candidate paths at 3 to keep CODEOWNERS lookup focused — the most-mentioned path wins ties.