CI Refactoring Blog — Channeltalk Reference Application: Single Comprehensive Post

Sprint 107 — CI Refactoring Blog: Channeltalk Reference Application

Background

Sprints 102–106 refactored AlgoSu's CI pipeline over 5 sprints using the Channeltalk Backend CI Refactoring post (https://channel.io/ko/team/blog/articles/backend-ci-refactoring-73fca77d) as a reference. Decisions, numbers, and lessons remain in docs/adr/sprints/sprint-102.md ~ sprint-106.md and memory/project-ci-refactoring-roadmap.md, but were scattered without a reader-facing narrative.

Sprint 107 consolidates this 5-sprint journey into a single comprehensive blog post. The differentiator is centering the narrative on "the process of translating — not directly translating — the reference into the context of a solo developer + AI agent orchestration."

Goals

ItemContentStatus
Single comprehensive postChanneltalk reference application frame, 500–700 lines✅ Complete (501 lines)
Fact verificationDirect cross-reference with Sprint 102–106 ADR numbers✅ Complete (13 items matched)
Blog build verificationblog/ npm run build SSG 20/20✅ Complete
Deployment (main merge)PR squash merge✅ Complete (#128, #129)

Decisions

D1. Structure: Single Comprehensive Post (not a series)

Decision: Single post, not a 4–5 part series.

Rationale:

  • The reference Channeltalk post itself is a single post format
  • The central message spanning all 5 sprints ("reference → context translation → 0 implementation lines conclusion") is weakened when split into series
  • User blog workflow feedback: "single-post cycle (plan→review→write)" — single post is natural

Alternative deferred: Series expansion retained as Sprint 108+ candidate but not mandatory.

D2. Frame: "Reference Application"

Decision: Narrative framed as "borrow principles + translate context."

Rationale:

  • Existing blog post cicd-ai-guardrails.mdx covers CI topic → need differentiated angle to avoid duplication
  • Not simply translating/summarizing the Channeltalk post, but explicitly comparing what was and wasn't adopted provides reader value
  • Highlights AlgoSu's unique context of solo developer + AI agent structure

D3. Agent Assignment Correction — Scribe Alone

Decision: Simplified plan's original Librarian/Herald assignment to Scribe alone.

Rationale:

  • Librarian: Role is DB schema/TypeORM Migration management — unsuitable for blog fact verification
  • Herald: Role is Frontend UI implementation — commit/PR role is Oracle/PM domain
  • Scribe: Dedicated to documents/memory/prompts → can handle fact verification (direct ADR Read) + MDX draft writing in parallel

Lesson: Agent domain matching must be re-verified by Oracle immediately before dispatch, separate from plan approval. Extension of Sprint 102 "CI is Gatekeeper's domain → Architect reassignment" lesson.

D4. Title Simplification

Decision: Draft title "I read the Channeltalk CI post and did 5-sprint refactoring myself" → "AlgoSu CI Refactoring".

Rationale: User feedback. Clean and short form. Reference mention sufficiently handled in post intro — removed from title redundancy.

D5. "We" → "AlgoSu" Tone Unification

Decision: Replace all 4 occurrences of "we" (inappropriate for solo developer context) with "AlgoSu."

Rationale: AlgoSu is solo developer + AI agent structure, so "we" is ambiguous. Unified with proper noun subject.

Outputs

  • New file: blog/content/posts/ci-refactoring-reference-to-reality.mdx (501 lines)
    • frontmatter: title="AlgoSu CI Refactoring", category=journey, order=8, tags=["ci-cd", "github-actions", "ai-dev", "refactoring"]
    • 9 sections: intro → reference rationale → 4-sprint roadmap → S102 → S103-104 → S105 → S106 → 4 principles → unfinished story
    • MDX components: <PhaseTimeline>, <PhaseMilestone>, <Callout> (info/warn/success), <MetricGrid>, <MetricCard>
    • 6 code blocks: dependabot.yml, auto-merge workflow YAML, Branch Protection, Composite action YAML, check-coverage.mjs, commitlint + husky setup
    • Channeltalk URL cited 2 times
  • PR #128 (599a71f): 501 lines new file
  • PR #129 (a3ab0b4): Title simplification (1-line edit)
  • Verification: blog/ npm run build SSG 20/20 success, /posts/ci-refactoring-reference-to-reality path renders correctly

Fact Verification Results (Scribe cross-reference, 13 items)

ItemADR sourceValue used
Dependabot pending PRssprint-10230 → 2
PR #102 SHAsprint-10246aeb73
Composite lines deletedsprint-10467 lines
3 Post run numberssprint-10524702740418 · 24702828670 · 24703075569
Quality Pre→Postsprint-105+0.4% (+0.1s)
Audit Pre→Postsprint-105-8.9% (-1.8s)
Test Pre→Postsprint-105+3.9% (+0.8s)
runner-minutes savingssprint-10575%
Frontend branches Beforesprint-10669.55% (1302/1872)
Frontend branches Aftersprint-10676.42%
New test countsprint-10677
PR #121·#122sprint-106Match
Channeltalk URLplan + all ADRsAccurate

All 13 items ✅ matched.

Lessons Learned

1. Effectiveness of Blog Workflow Feedback

The "single-post cycle (plan→review→write)" feedback dominated Sprint 107's entire flow. User agreement on single/series choice at plan stage → one tone adjustment ("we→AlgoSu") after draft → one title simplification revision cycle. If executed in batch, both would have required post-production correction at higher rework cost.

2. Re-verification Timing for Agent Domain Matching

Librarian/Herald were assigned in the plan, but Oracle re-verified agent specs immediately before dispatch and corrected to Scribe alone. Reconfirmed that a "domain matching re-check" step is needed between plan approval and dispatch. Same lesson from Sprint 102 establishing as an operating rule.

3. Importance of Direct ADR Number Cross-Reference

Blog posts are public-facing, so factual errors damage credibility. Scribe's approach of directly Reading 5 ADRs and including a 13-item cross-reference table in the results report greatly reduced post-publication verification burden. Consider promoting "agent self-cross-reference table" pattern as a standard for blog/public document work.

4. MDX Component Reuse for Tone Consistency

Reused <PhaseTimeline>, <Callout>, <MetricGrid> usage patterns from the existing cicd-ai-guardrails.mdx to maintain blog-wide tone consistency. Pure reuse was sufficient without creating new components.

5. Benefit of "No Commit Until Review" Constraint

The "no commit after draft, wait for user review" constraint aligned precisely with the blog workflow feedback. User revisions (tone adjustment → title) intervened at the point when Scribe's result report arrived in inbox, naturally separating into 2 distinct PRs.

Carried Over (Sprint 108+)

Carry-overs Inherited from Sprint 106 (No Change)

  • Sprint 108 core decision candidates: Blog/Frontend host-side build migration (true L2 achievement path, 40–60% expected reduction)
  • APK_CACHE_BUST conditionalization, NestJS tsc incremental, Monaco dynamic import, heavy deps audit
  • ai-analysis branch=true, submission/problem/identity lcov actual collection
  • check-coverage.mjs per-service independent gate

New Carry-overs from Sprint 107

  • Series expansion candidate: Currently 1 single comprehensive post. Deep-dive on specific topics (pre-consultation pattern, Composite Action, Coverage Gate, etc.) not mandatory but retained as Sprint 108+ option.
  • Blog post metadata policy: order field manually managed (max+1 method). As posts increase, consider switching to auto-assignment or date-based sorting.
  • blog/content/posts/ci-refactoring-reference-to-reality.mdx — output
  • blog/content/posts/cicd-ai-guardrails.mdx — tone/component reference
  • docs/adr/sprints/sprint-102.md ~ sprint-106.md — fact sources
  • memory/project-ci-refactoring-roadmap.md — full roadmap
  • /Users/leokim/.claude/plans/drifting-wishing-thacker.md — approved plan
  • External: Channeltalk Backend CI Refactoring