CCT-AIDLC Process Flow

The CCT-tailored AI-DLC variant — a streamlined Inception + Construction flow for lifting one legacy WinForms artifact into Insight Cash. Companion to process_flow.md. Operations phase is deferred (out of scope for now).

Two phases, eleven steps (plus one optional bridge step — Step 2.5), two review gates. Step 2 (drop sources) is the only manual step. In Construction, Steps 5–8 run in parallel from the same Inception artifacts, then converge on Step 9 (bolts). The single review gate at the end of design covers all four parallel artifacts plus bolts.md together.

Phases

  • Phase 1 — Inception (Steps 1–4)
  • Phase 2 — Construction (Steps 5–11)
  • Phase 3 — Operations (deferred)

Reviewer role (node fill color)

  • Author — author judgment, no formal reviewer at step
  • PO — Product Owner
  • TL — Tech Lead
  • AR — Architect
  • QA — Quality Assurance
  • Dev — Developer (peer)

Color reflects the primary reviewer at the gate that covers this step. Multiple reviewers are listed in each node's body.

Reading the diagram

  • Each node is a step from process_flow.md.
  • Bold lines = sequential flow; double-line subgraphs = parallel sections.
  • 🛠️ Skill lines name the /cct-* skill that produces the artifact. The 11 step skills (one optional — /cct-intent-update) live at .claude/skills/cct-*/SKILL.md.
  • Dashed edges (-.->) mark the optional path to Step 2.5; solid double-lines (==>) are the standard sequential flow.
  • All artifact paths shown are relative to cct-aidlc/<intent>/.
  • Diamond nodes (◇) are the two formal review gates — the only places where the backlog row's Status flips.
flowchart TD

  subgraph P1["Phase 1 — Inception"]
    direction TB
    S1["Step 1 — Scaffold the intent
📄 inception/intent.md (skeleton)
📄 inception/prompts.md (P-1)
📁 inception/sources/ (empty)
📁 construction/ (empty)
📋 New row in intent-backlog.md (Draft)
🛠️ Skill: /cct-intent <intent>
Author: AI scaffolder · Reviewer: author judgment"] S2["Step 2 — Drop sources (manual)
📁 inception/sources/ (legacy XML, screenshots, SP DDL, table DDL, vendor specs)
📄 P-N entry in inception/prompts.md
🛠️ Skill: none — manual file copy
Author: author (person) · Reviewer: author judgment"] S2H["Step 2.5 — Targeted intent edit (optional)
📄 revisions to inception/intent.md (specific fields)
📄 P-N entry in inception/prompts.md
🛠️ Skill: /cct-intent-update <intent> <text> (single-shot)
Locks in Pathway / Depends on / Goal / Success criterion /
Constraints / Deviations before Q&A opens.
Skip when nothing to lock in.
Author: AI applies stated edits · Reviewer: author judgment"] S3["Step 3 — Build the plan (multi-turn Q&A)
📄 inception/plan.md (with full Q&A trail)
📄 refinements to inception/intent.md
📄 P-N entry in inception/prompts.md
🛠️ Skill: /cct-plan <intent> (conversational)
Author: AI proposes · Reviewer: PO at end-of-Inception gate"] S4["Step 4 — Elicit user stories
📄 inception/user_stories.md
📄 P-N entry in inception/prompts.md
🛠️ Skill: /cct-stories <intent>
Author: AI · Reviewer: PO at end-of-Inception gate"] G1{"◇ End-of-Inception review gate
PO (always) + TL/AR (conditional — only if plan introduces
new shared component or new architecture)
Outcome on approve: backlog flips Draft → Approved
(Construction unlocks)"} end subgraph P2["Phase 2 — Construction"] direction TB subgraph PARA["Steps 5–8 (in parallel from approved Inception artifacts)"] direction LR S5["Step 5 — Domain model
📄 construction/domain.md
aggregates · numbered invariants (DM I1, DM I2, …) ·
state machines · events · policies
🛠️ Skill: /cct-domain <intent>
Skip for report pathway
Reviewer: AR + TL at end-of-design"] S6["Step 6 — ADRs / TDRs
📄 construction/adrs.md (feature-local)
📄 architecture/adrs/ADR-NNN-*.md (cross-cutting promotions)
🛠️ Skill: /cct-adrs <intent>
Skip for report pathway unless cross-cutting
Reviewer: AR + TL at end-of-design"] S7["Step 7 — Detailed design
📄 construction/design.md
DB summary · IaC summary · component map · routes ·
zod schemas · SP signatures · CLI surface · error mapping
🛠️ Skill: /cct-design <intent>
Reviewer: TL + AR at end-of-design"] S8["Step 8 — Test plan
📄 construction/tests.md
T-N IDs · AC refs · buckets (unit/integration/contract/smoke/e2e) ·
standards-conformance tests
🛠️ Skill: /cct-tests <intent>
Reviewer: QA + TL at end-of-design"] end S9["Step 9 — Bolt decomposition
📄 construction/bolts.md
B-1..B-N deployable slices · files created/modified ·
depends-on DAG · T-N tests covered · verification commands
🛠️ Skill: /cct-bolts <intent>
Author: AI · Reviewer: TL at end-of-design"] G2{"◇ End-of-design review gate
TL + AR + QA (always) · PO (only if scope/UX shifted from Inception)
Outcome on approve: backlog flips Approved → In-Construction
(Step 10 unlocks)"} S10["Step 10 — Per-bolt code
💻 Code + T-N tests, one commit per bolt
📄 P-N entry in construction/prompts.md (per bolt)
🛠️ Skill: /cct-code-bolt <intent> <B-id>
One invocation per bolt, in dependency order
Author: AI + driver Dev · Reviewer: TL or peer Dev"] S11["Step 11 — Doc sync
📄 architecture/api_reference.md (rows added)
📄 architecture/cli_reference.md (rows added)
📄 architecture/cct_cli_user_guide.md (recipes added)
📄 (sometimes) feature-specific guide updates
🛠️ Skill: /cct-doc-sync <intent>
Mechanical sync — does NOT introduce new design decisions
Reviewer: AR (or TL) — light review
Outcome on approve: author runs /cct-backlog --update <intent> In-Operations to flip the row"] end subgraph P3["Phase 3 — Operations (deferred)"] direction TB Sunset["In-Operations
Live in production · Operations sub-states TBD
Skills for runbooks / monitoring / incident response
are deferred to a later spec pass.
"] end S1 ==> S2 S2 -.->|optional| S2H S2 ==> S3 S2H ==> S3 S3 ==> S4 S4 ==> G1 G1 ==> S5 G1 ==> S6 G1 ==> S7 G1 ==> S8 S5 ==> S9 S6 ==> S9 S7 ==> S9 S8 ==> S9 S9 ==> G2 G2 ==> S10 S10 ==> S11 S11 ==> Sunset classDef rAuthor fill:#f0f0ec,stroke:#888,stroke-width:1.5px,color:#1a1a1a classDef rPO fill:#cfe2f3,stroke:#4a78c9,stroke-width:1.5px,color:#1a1a1a classDef rTL fill:#d9c7e8,stroke:#7e5da8,stroke-width:1.5px,color:#1a1a1a classDef rAR fill:#e8d4f4,stroke:#715197,stroke-width:1.5px,color:#1a1a1a classDef rQA fill:#fff3a8,stroke:#b89b1f,stroke-width:1.5px,color:#1a1a1a classDef rDev fill:#c8e6c9,stroke:#4d8a59,stroke-width:1.5px,color:#1a1a1a classDef gate fill:#f4c7c3,stroke:#b8483f,stroke-width:2px,color:#1a1a1a class S1,S2,S2H rAuthor class S3,S4 rPO class S5,S6 rAR class S7 rTL class S8 rQA class S9,S10 rTL class S11 rAR class Sunset rAuthor class G1,G2 gate style P1 fill:#f4f8ff,stroke:#4a78c9,stroke-width:2px,color:#1a3a6e style P2 fill:#fffaf0,stroke:#c98a2c,stroke-width:2px,color:#6e4a1a style P3 fill:#f0faf3,stroke:#3d8d57,stroke-width:2px,color:#1a4a2c style PARA fill:#fff,stroke:#aaa,stroke-width:1.5px,stroke-dasharray:4 3,color:#555

Skill inventory — at a glance

Eleven step skills implement the workflow (one — /cct-intent-update — is optional and runs only when the author wants to lock in specifics before Q&A), plus one CCT-AIDLC workspace-utility skill (/cct-backlog), six branching-utility skills (/cct-pr-inception, /cct-pr-construction, /cct-chore, /cct-release, /cct-rebase-intent, /cct-commit) operationalizing the git workflow in cct-aidlc/branching.md, and two repo-wide read-only utility skills (/change-summary, /test-summary). Step 2 (drop sources) is the only manual step. /cct-plan is the only multi-turn step skill; /cct-code-bolt is the only step skill with two arguments. Every step skill except /cct-intent refreshes the Last Updated column on the affected intent's backlog row; Status flips happen at the manual review gates via /cct-backlog --update, not inside step-skill runs.

Phase Skill Step Notes
Inception/cct-intent <intent>1 — ScaffoldCreates folder + intent.md skeleton + prompts.md + Draft row in intent-backlog.md.
Inception/cct-intent-update <intent> <text>2.5 — Targeted intent edit (optional)Single-shot bridge between /cct-intent and /cct-plan. Reads intent.md + sources/ and applies the author's plain-language instructions to specific fields (Goal, Success criterion, Pathway, Depends on, Constraints, Deviations, Source attribution). Does not start a Q&A loop, does not draft plan.md, does not flip Status. Skip when there's nothing to lock in before Step 3.
Inception/cct-plan <intent>3 — Build the planMulti-turn, conversational. Q&A loop until ambiguities resolved; refines intent.md in place.
Inception/cct-stories <intent>4 — Elicit storiesSingle-shot. One story per legacy interaction. Standard CCT persona cast.
Construction/cct-domain <intent>5 — Domain modelParallel with Steps 6, 7, 8. Skips for report pathway.
Construction/cct-adrs <intent>6 — ADRs / TDRsParallel with Steps 5, 7, 8. Skips for report pathway unless cross-cutting.
Construction/cct-design <intent>7 — Detailed designParallel with Steps 5, 6, 8. Pathway-shaped (thinner for report).
Construction/cct-tests <intent>8 — Test planParallel with Steps 5, 6, 7. Authored from Inception artifacts only.
Construction/cct-bolts <intent>9 — Bolt decompositionAfter Steps 5–8 are author-ready. Decomposes design into B-1..B-N.
Construction/cct-code-bolt <intent> <B-id>10 — Per-bolt codeSequential per bolt, in dependency order. One commit per bolt.
Construction/cct-doc-sync <intent>11 — Doc syncAfter all bolts shipped. Mechanical sync to architecture/ reference docs.
Workspace utility/cct-backlog [--update <intent> <status>]cross-stepReads the backlog table; --update flips Status at the manual gates (Draft → Approved, Approved → In-Construction, In-Construction → In-Operations, In-Operations → Sunset) and appends a YYYY-MM-DD HH:MM-stamped P-N entry to the appropriate prompts.md. Status flips happen only through this skill, never from a step skill.
Branching utility/cct-pr-inception <intent>End-of-Inception gateDrafts Inception PR (intent/<slug>/inceptionintent/<slug>). Verifies the three Inception artifacts and backlog Status is Draft. Emits gh pr create --base intent/<slug> --head intent/<slug>/inception and post-merge construction-branch setup commands for the user.
Branching utility/cct-pr-construction <intent>End-of-design gateDrafts Construction PR (intent/<slug>/constructionintent/<slug>, long-lived). Verifies pathway-aware Construction artifacts and backlog Status is Approved. Emits gh pr create --draft --base intent/<slug> and the post-merge gh pr ready command for the user.
Branching utility/cct-chore <slug>cross-stepCuts chore/<slug> off origin/develop for administrative work. Branch-create only — leaves any working-tree changes uncommitted for the user (typically committed via /cct-commit).
Branching utility/cct-release [vN.M]Release eventAdvisory only. Runs read-only release safety checks and emits the git checkout main && git merge --ff-only develop && git tag && git push sequence. Never touches main.
Branching utility/cct-rebase-intent <intent>cross-stepRebases an open intent/<slug> onto updated origin/develop. Auto-fetches and auto-runs git rebase. On clean rebase, emits git push --force-with-lease. On conflict, halts and emits resolution commands. Never pushes.
Branching utility/cct-commit [subject]cross-stepCommits the currently-staged changes locally with a generated branch-aware message (subject prefix from intent/<slug>/inception, intent/<slug>/construction, intent/<slug>, issue/ISS-NNN-*, chore/<slug>, or Conventional Commits). Refuses on main / develop. Never stages, pushes, or opens PRs.
Repo utility/change-summary [days]cross-stepTwo-phase summary of recent git activity (default 7-day window). Phase 1 prints commit subjects grouped by author with the current user first; Phase 2 expands diffs for entries the user picks. Read-only — no files written. Useful for catching up after time away or pre-reading recent activity before End-of-Inception / End-of-design reviews.
Repo utility/test-summary [--detailed] [path-or-tool]cross-stepTwo-axis test inventory — rows are CCT-AIDLC test types (unit / integration / contract / smoke / e2e), columns are areas (frontend / handlers / CLI / cross-cutting), cells show counts. --detailed lists file paths under each non-zero bucket; the optional positional argument filters by path or tool/intent name. Read-only — no files written. Useful during Step 8 (/cct-tests) to validate tests.md bucketing and surface coverage gaps.

Repo-wide utility skills

Two read-only utility skills sit alongside the CCT-AIDLC step skills. Neither is part of the workflow — they don't read or write any cct-aidlc/<intent>/ artifacts and don't refresh the backlog — but both come up often enough during CCT-AIDLC work that they're worth keeping next to the canonical inventory. SKILL.md files live at .claude/skills/change-summary/SKILL.md and .claude/skills/test-summary/SKILL.md.

/change-summary [days]

Repo utility — recent git activity

Two-phase summary of git activity over a days-wide window (default 7). Phase 1 lists every commit on main and on other remote branches in the window, grouped by author with the current user listed first and tagged (you). Subjects only — clearly-related consecutive commits by the same author are consolidated into a single themed entry. Phase 2 reads the actual diff for entries the user picks and writes a richer synopsis. The header echoes the absolute window dates so the output is reproducible.

Argument
Optional integer = window size in days (default 7).
Output
Printed to chat — no files written.
When useful
Catching up after time away · drafting status updates · populating PR descriptions · pre-reading recent activity before an End-of-Inception or End-of-design review.
SKILL.md
.claude/skills/change-summary/SKILL.md

/test-summary [--detailed] [path-or-tool]

Repo utility — test inventory

Two-axis test inventory across the repo. Rows are CCT-AIDLC test types: unit, integration, contract, smoke, e2e — same buckets /cct-tests uses in tests.md, so the inventory cross-references cleanly with any intent's test plan. Columns are areas: frontend, handlers, CLI, cross-cutting. Cells show test-case counts. The skill is read-only — it discovers files via find and counts cases via grep. No code is built, no tests are run, no files are written.

Flags
--detailed (or --files) lists file paths under each non-zero bucket. Default is counts-only.
Filter
Optional positional arg — case-insensitive substring filter against discovered file paths (department-maintenance, app/cli/test/unit).
Output
Printed to chat — no files written.
When useful
During Step 8 (/cct-tests) to validate tests.md bucketing against the existing repo pattern · coverage audit before End-of-design review · finding gaps (e.g., a tool that ships handlers but no contract test).
SKILL.md
.claude/skills/test-summary/SKILL.md