/triage

name: triage
description: "Periodic-aggregation root-cause triage of the open auto-improvement queue"
argument-hint: "[--auto-improvement] [--repo OWNER/REPO] [--limit N] [--include-fp-acknowledged] [--json]"
user-invocable: true
user_facing: true
allowed-tools: [Bash, Read]

Triage: Periodic-Aggregation Root-Cause Triage

`/triage --auto-improvement` reads the open issues labelled `auto-improvement`, clusters them by root cause, surfaces cross-cluster dependencies, and emits a ranked work queue.

This command is the periodic-aggregation counterpart to per-session `cia.md` issue filing. CIA files one issue per session finding; this command groups the accumulation by root cause so a maintainer can drain the queue weekly.

Implementation

ARGUMENTS: {{ARGUMENTS}}

STEP 0: Parse arguments

Recognized flags:

• `--auto-improvement`: Triage the `auto-improvement`-labelled queue (currently the only

supported mode).

• `--repo OWNER/REPO`: GitHub repository (default: `akaszubski/autonomous-dev`).
• `--limit N`: Maximum issues to fetch (default: 200). If the result hits the limit, a

WARN line is printed to stderr.

• `--include-fp-acknowledged`: Include issues labelled `fp-acknowledged` (filtered out

by default — these have been triaged as false positives).

• `--json`: Emit machine-readable JSON instead of the human-readable text format.

If no flags are passed, default to `--auto-improvement` (the only mode).

STEP 1: Invoke the analyzer

python3 -c "
import sys, os
for _p in ('.claude/lib', 'plugins/autonomous-dev/lib', os.path.expanduser('~/.claude/lib')):
    if os.path.isdir(_p):
        sys.path.insert(0, _p)
        break
from issue_triage_analyzer import main
sys.exit(main({{FLAGS_AS_ARG_LIST}}))
"

Replace `{{FLAGS_AS_ARG_LIST}}` with the parsed flag list (e.g. `['--auto-improvement', '--repo', 'owner/repo']`). If `--json` was passed, the analyzer emits JSON on stdout; otherwise it emits a human-readable report.

STEP 2: Display findings

Display the analyzer's stdout verbatim. The default human-readable form groups findings by `[root_cause_tag] [sub_cluster_id]` with ranking score, severity, cluster size, and member issue numbers. Cross-cluster dependencies (clusters sharing a file path) are noted inline. A `--json` invocation returns the same data as a deterministic JSON array suitable for piping to other tools.

Idempotence

On a clean queue with unchanged contents the command produces byte-identical output across runs. Sorts and tie-breakers are total: rank score DESC, then root cause tag ASC, then sub-cluster ID ASC, then issue numbers within a cluster ASC. The only time-varying input is `now` (used for recency decay).

When to run

Weekly, or after a CIA-heavy session that filed many `auto-improvement` issues. The output is a work queue, not a destructive action. Pair with `gh issue list --label auto-improvement` to spot-check.

Filing aggregates

The daily aggregate lifecycle helper `plugins/autonomous-dev/lib/daily_aggregate_manager.py::open_or_supersede_daily_aggregate` is the only sanctioned path for creating aggregate issues with guarded titles.

Direct `gh issue create` commands with titles starting with "Auto-triage findings —" or "[CRITICAL] AI triage —" are HARD-BLOCKED by `unified_pre_tool.py::_detect_daily_aggregate_direct_filing`.

Living-aggregate semantics:

• Idempotent within same UTC day: repeated calls edit the existing aggregate
• Supersedes prior day aggregates: closes old aggregate with "Superseded by #NEW" comment
• Only the triage-aggregate marker command can bypass the title guard

This ensures aggregate issues follow consistent lifecycle patterns and prevent duplicate filing.