Workflow Completeness: self-development README.md has three additional inaccuracies post-#890

[kelos-bot]

🤖 Kelos Self-Update Agent @gjkim42

Area: Workflow Completeness

Summary

self-development/README.md has three factual inaccuracies not covered by #890. Two were introduced by changes merged after #890 was filed (2026-04-03), and one is a pre-existing error #890 missed. These are separate from #890's five findings and should be fixed independently.

Findings

1. api-reviewer missing from inline AgentConfig list (line 9)

Line 9 lists agents with inline AgentConfigs:

"others (workers, planner, fake-user, fake-strategist, self-update, image-update) define their own AgentConfig inline"

This omits both reviewer (#890 Finding 2 covers this) and api-reviewer (NOT covered by #890). The kelos-api-reviewer was added in commit ba319af (2026-04-11), which updated the README's TaskSpawner table and added a full ### kelos-api-reviewer.yaml section — but did not update line 9.

Evidence: kelos-api-reviewer.yaml defines its own kelos-api-reviewer-agent AgentConfig inline, matching the pattern of all other inline configs. The README table (line 18) correctly lists api-reviewer but line 9 does not.

2. Triage trigger description includes removed labeled action (lines 20, 154, 158)

The README describes the triage trigger in three places:

• Line 20: Webhook: issue opened/labeled/reopened (needs-actor)
• Line 154: Picks up open GitHub issues labeled needs-actor and performs automated triage.
• Line 158: GitHub issue opened/labeled/reopened webhooks with needs-actor

Actual triage filters (from kelos-triage.yaml lines 12–22):

• issues.opened — triggers on ANY newly opened issue (no needs-actor required, only excludes triage-accepted)
• issues.reopened — triggers only with needs-actor label

Two errors:

1. The labeled action filter was removed in commit 1ab8a7b ("remove labeled filters that cause duplicate task spawns"). The README still lists it.
2. The opened action does not require needs-actor — this was changed in commit 21bdf27 ("trigger triage on issue opened without needs-actor"). The README implies all actions require needs-actor.

Correct description: "Webhook: issue opened, or issue reopened with needs-actor"

3. Reviewer section claims agent runs make test (line 88)

The README's reviewer key features list states:

• Runs make test to verify tests pass

The actual reviewer prompt (kelos-reviewer.yaml) explicitly prohibits running validation:

• Line 130–131: "Review test adequacy from the diff and surrounding code; do not rerun validation as part of the review"
• Line 264: "Do NOT run local validation commands or wait for CI status as part of the review"

The reviewer is a read-only agent that reviews code statically. It does not and should not run make test. This predates #890 but was not included in its findings.

Impact: Users adapting the reviewer pattern would incorrectly add make test to their review agent's workflow, which contradicts the read-only design and wastes task execution time.

Proposed Fix

1. Add api-reviewer to line 9's inline AgentConfig list (alongside reviewer per Workflow Completeness: self-development README.md has five factual inaccuracies against actual TaskSpawner definitions #890 Finding 2)
2. Update triage trigger descriptions on lines 20, 154, and 158 to: "issue opened, or issue reopened with needs-actor"
3. Replace the make test bullet on line 88 with: "Reviews test adequacy from the diff — does not run validation commands"

Not covered by existing issues

• Workflow Completeness: self-development README.md has five factual inaccuracies against actual TaskSpawner definitions #890 covers 5 other README inaccuracies but not these three (api-reviewer didn't exist when Workflow Completeness: self-development README.md has five factual inaccuracies against actual TaskSpawner definitions #890 was filed; triage filters were changed after Workflow Completeness: self-development README.md has five factual inaccuracies against actual TaskSpawner definitions #890; reviewer make test claim was missed)
• Prompt Tuning: kelos-reviewer convention checklist incomplete and missing make verify step #904 covers the reviewer's convention checklist gaps but not the README's false make test claim

[kelos-bot]

🤖 Kelos Agent @gjkim42

Kelos Triage Report

Kind: kind/docs
Priority: priority/important-longterm — Documentation inaccuracies that could mislead users adapting these patterns, but not blocking adoption
Actor: actor/kelos — All three fixes are concrete text changes to a single README with objectively verifiable acceptance criteria

Status: STILL VALID

Evidence:

• Finding 1 (api-reviewer missing from line 9): Confirmed. self-development/README.md line 9 lists inline AgentConfig agents as (workers, planner, fake-user, fake-strategist, self-update, image-update) but omits api-reviewer. Verified that kelos-api-reviewer.yaml defines its own kelos-api-reviewer-agent AgentConfig inline. Separate from Workflow Completeness: self-development README.md has five factual inaccuracies against actual TaskSpawner definitions #890 Finding 2 which covers reviewer being missing from the same line.
• Finding 2 (triage trigger description): Confirmed. Lines 20, 154, and 158 still reference labeled action and imply all triggers require needs-actor. Verified in kelos-triage.yaml: only two filters exist — issues.opened (no needs-actor required, only excludes triage-accepted) and issues.reopened (requires needs-actor). No issues.labeled filter exists.
• Finding 3 (reviewer make test claim): Confirmed. Line 88 states "Runs make test to verify tests pass" but the actual kelos-reviewer.yaml prompt explicitly prohibits this at lines 130–131 ("do not rerun validation as part of the review") and line 264 ("Do NOT run local validation commands"). The reviewer is a read-only static analysis agent.

Duplicates found: #890 is related but not a duplicate — it covers 5 different README inaccuracies. Finding 1 here touches the same line 9 as #890 Finding 2, but identifies a separate omission (api-reviewer vs reviewer). #904 covers reviewer prompt gaps, not the README's false make test claim.

Recommendation: KEEP OPEN