feat: 변경 식별자 소비처 탐색·관찰 가능 지적 QA 편입·QA 사전점검과 재실행 스크립트 (v2.4.0)

README.ko.md

@@ -82,7 +82,7 @@ $EDITOR .sdlc/config.md                          # build, test, lint, run, e2e,
   PLAN.md       한 장: 이해, AS-IS → TO-BE, 부작용, 변경, 시나리오
   review.md     리뷰 라운드, 각 라운드는 VERDICT로 끝남
   evidence.md   검사, 시나리오별 결과 한 행, 납품 기록
-  logs/         모든 PASS·FAIL 뒤의 원본 출력
+  logs/         모든 PASS·FAIL 뒤의 원본 출력, scripts/는 시나리오별 재실행 스크립트
 ```
 
 끝난 기능은 `.sdlc/archive/`로 옮겨집니다. 검증된 사실은 `knowledge` 폴더로

README.md

@@ -83,7 +83,7 @@ always says what comes next.
   PLAN.md       one page: understanding, AS-IS → TO-BE, side effects, changes, scenarios
   review.md     reviewer rounds, each ending in VERDICT
   evidence.md   checks, one result row per scenario, delivery
-  logs/         raw output behind every PASS and FAIL
+  logs/         raw output behind every PASS and FAIL; scripts/ re-runs each scenario
 ```
 
 Closed features move to `.sdlc/archive/`. Verified facts move to the

VERSION

@@ -1 +1 @@
-2.3.0
+2.4.0

log/changelog-v2.4.0.md

@@ -0,0 +1,41 @@
+# v2.4.0 — find consumers before QA does, cheaper QA rounds
+
+Lessons from one feature (a front-end route restructure) that took three
+review rounds and three QA rounds. Each QA round re-drove 18 to 20 browser
+scenarios by hand (45 and 38 minutes); two of its three code defects were
+knowable before QA. No script change: the rules moved.
+
+## Changes
+
+- **Consumers of changed identifiers.** The change turned list routes into
+  redirects. A privacy-audit logger still compared the old route names, so
+  its log stopped silently; the plan missed it and the first review called
+  the old references harmless. The logic scout now lists every consumer of
+  an identifier the change renames, retires or redirects — searching the
+  literal value too, and counting consumers that only compare it (loggers,
+  audit and analytics events, permission checks, watchers, flags, links from
+  other apps). PLAN.md's side effects cover each; the reviewer blocks a plan
+  that skips them and a diff that leaves one comparing the old value.
+- **Observable review findings become scenarios.** A finding the first
+  review rated non-blocking (a redirect missing a required parameter) was
+  only driven in QA round 2, where it failed and cost a loop. A non-blocking
+  finding a screen, API, job or log could show now ends in `(observable)`,
+  and QA runs each as a scenario in the first round.
+- **QA preflight.** Three environment faults were each found mid-run: a dev
+  proxy pointing at a local API that was not running, a dev server killed by
+  its host's 30-minute background limit, and a browser driver that cannot
+  start in a sandboxed peer agent. QA now checks once, before any scenario,
+  that one real call per route reaches the intended environment, that the
+  driver starts, and that long-running processes outlive the run, and records
+  what it learned.
+- **Re-runnable scenarios, cheaper rounds.** QA saves the script behind each
+  scenario in `logs/scripts/`. A later round re-runs every script against
+  the new commit and drives by hand only FAIL and BLOCKED rows, new rows and
+  rows the new diff touches, writing `logs/r<round>-<ID>.txt` beside the old
+  logs. Every row still gets a run on the new commit. At delivery, scripts
+  worth keeping go to the project's `e2e` suite or into Delivery as a
+  follow-up.
+- **A code FAIL is a bug fix.** Back from Evidence, each code FAIL gets its
+  regression test first, failing for the reported reason, as Implement
+  already asks of bug fixes. In that feature one fix went to review without
+  one and the review failed on exactly that.

phases/2-plan.md

@@ -15,7 +15,7 @@ writing one file in `refs/`:
 |---|---|---|
 | source | `refs/source.md` | the request's text, the policies and canonical docs it touches, earlier attempts and reverts and why |
 | data | `refs/data.md` | data models and fields, who writes and reads each, the states existing records are in |
-| logic | `refs/logic.md` | entry points, call chains, every caller in every tier and repository the change crosses, shared code |
+| logic | `refs/logic.md` | entry points, call chains, every caller in every tier and repository the change crosses, shared code, every consumer of an identifier the change renames, retires or redirects |
 | flow | `refs/flow.md` | how each persona reaches the behavior today, observed by running it, with the result |
 
 A bug adds `refs/repro.md`: the failure reproduced — as a failing test where
@@ -35,7 +35,8 @@ evidence. Link, do not copy.
   TO-BE is `unchanged`, with a scenario.
 - **Side effects** — every other reader or writer of the touched data, every
   caller that sends today's shapes (missing, empty, null, old values),
-  records made before the change, shared code — each with how it stays
+  records made before the change, shared code, every consumer of a changed
+  identifier (even one that only compares it) — each with how it stays
   correct and the scenario that checks it. Then the risk flags line.
 - **Changes** — files and symbols, in an order that keeps checks passing;
   the smallest runnable path first.

phases/3-build.md

@@ -60,4 +60,6 @@ the loop: show the human the trail; only `sdlc.sh approve plan <slug>
 ## Back from Evidence
 
 `sdlc.sh next` names the scenarios that failed for code reasons; evidence.md
-and `logs/` hold what was seen. Fix them, commit, and review again.
+and `logs/` hold what was seen. Each is a bug fix: write its regression test
+first and watch it fail for the reported reason, then fix, commit, and review
+again.

phases/4-evidence.md

@@ -9,7 +9,10 @@ Dispatch QA (`roles/qa.md`) with intent.md, PLAN.md, `refs/`, config.md and
 the reviewed commit. QA runs the checks and every scenario, saves raw output
 in `logs/`, adds scenarios for anything new it finds, and fills `evidence.md`
 (`templates/evidence.md`): one row per scenario — PASS, FAIL with a cause,
-or BLOCKED with what is missing. Scenarios are never dropped: `sdlc.sh next`
+or BLOCKED with what is missing. QA preflights the environment once, runs each
+`(observable)` review finding as a scenario, and saves the script behind
+every scenario, so a later round re-runs them all and drives by hand only
+what failed or what the new diff touches. Scenarios are never dropped: `sdlc.sh next`
 lists every ID the table needs, earlier QA-added ones included.
 A cause is `plan` when the plan was wrong or missed something (a caller, a
 data state, a rule, a scenario), and `code` when the plan was right and the
@@ -50,6 +53,8 @@ Delivery's `Policy:` line. Then:
 - move verified facts from the refs (and `refs/learnings.md`, if any) into
   `knowledge`: one place per fact, updating a page rather than adding a
   duplicate;
+- add the saved scripts (`logs/scripts/`) worth keeping to the project's
+  `e2e` suite, or name them as a follow-up in Delivery;
 - `sdlc.sh close <slug> shipped "<one line>"` — refused if the delivered
   tree differs from the verified one.
 

roles/qa.md

@@ -3,8 +3,15 @@
 Prove whether the change works for each persona through the real
 interfaces. Report what you observed; fix nothing.
 
-Inputs: intent.md, PLAN.md, `refs/`, config.md, and the commit(s) to verify.
+Inputs: intent.md, PLAN.md, `refs/`, review.md, config.md, and the
+commit(s) to verify.
 
+0. Preflight the environment once, before any scenario: the app answers;
+   one real call through each route the scenarios use reaches the intended
+   environment (proxy, host, account); the driver starts in this sandbox;
+   anything long-running is started detached with a limit longer than the
+   run. Fix what you may, or mark the scenarios it blocks BLOCKED with the
+   cause, and put what you learned in `refs/learnings.md`.
 1. Confirm the checked-out commit is the one you were given. Run `build`,
    `test`, `lint`, then `e2e` from config.md. A failure the base commit shows
    too is pre-existing: say so, with the base run.
@@ -13,20 +20,28 @@ Inputs: intent.md, PLAN.md, `refs/`, config.md, and the commit(s) to verify.
 3. For each PLAN.md scenario, as its persona: read the expected result first,
    then do the steps through the real interface — screen, API, CLI, job or
    library call. Read state back after every write. Save the raw output
-   (command, response, screenshot path, query result) to `logs/<ID>.txt`.
+   (command, response, screenshot path, query result) to `logs/<ID>.txt`,
+   and the script or command sequence that drove it to `logs/scripts/`, so
+   a later round re-runs it with one command.
 4. A scenario that cannot run here is BLOCKED, with what is missing. Never
    PASS by assumption; never a unit test in place of the real interface.
 5. Bug fix: run the regression test at the commit before the fix, in a
    disposable worktree (never stash or touch the human's tree): it must fail
    for the reported reason. Then run it after, and re-drive the surface where
    the bug was reported.
 6. Something the plan missed — another caller, a record state, a side
-   effect: add a row with the next free ID and run it. Besides your new
+   effect, an `(observable)` finding in review.md: add a row with the next
+   free ID and run it. Besides your new
    rows, the table holds exactly the IDs `sdlc.sh next <slug>` lists: a
    missing one is refused, and so is a row for a scenario the human removed.
    Write `\|` for a literal pipe inside a cell.
 7. Each FAIL gets a cause, with evidence: `plan` if the plan was wrong or
    missed something, `code` if the plan was right and the code is not.
+8. A later round (after a return to Build or Plan) re-runs every saved
+   script against the new commit; drive by hand only the FAIL and BLOCKED
+   rows, new rows, and rows whose surface `git diff <last verified
+   commit>..<new commit>` touches. Write the new output beside the old
+   (`logs/r<round>-<ID>.txt`); earlier logs stay.
 
 Fill `evidence.md` from `templates/evidence.md` and stop. If the feature
 folder is not writable from your sandbox (a symlink outside the allowed

roles/reviewer.md

@@ -9,15 +9,18 @@ Inputs: intent.md, PLAN.md, `refs/`. Blocking:
 - a caller, data reader or writer, record state or tier the refs show but
   PLAN.md ignores; a side effect without a scenario;
 - AS-IS stated without observation; a risk flag set to no that the changes
-  trip (migration, data deletion, public API, security, infra).
+  trip (migration, data deletion, public API, security, infra);
+- an identifier the plan renames, retires or redirects whose consumers
+  `refs/` do not list, or a listed consumer with no scenario.
 
 ## Diff mode (phase 3)
 
 Inputs: intent.md, PLAN.md, `refs/`, the commit or uncommitted diff
 (untracked files included). Read every changed file.
 Blocking:
 - it does not do what PLAN.md says, or does more;
-- a side effect PLAN.md names is left unhandled; input that callers send
+- a side effect PLAN.md names is left unhandled; a consumer that still
+  compares a renamed, retired or redirected value; input that callers send
   today is newly refused without intent.md authorizing it;
 - a policy changed with neither an update where it is written nor a
   recorded follow-up;
@@ -46,5 +49,7 @@ Append to `review.md`:
     - <what you read>
     VERDICT: PASS | FAIL
 
-FAIL when anything blocks. Every finding carries its evidence.
+FAIL when anything blocks. Every finding carries its evidence. End a
+non-blocking finding with `(observable)` when a screen, API, job or log
+could show it: QA runs each one as a scenario.
 Must not: write anything but `review.md`, or run mutating commands.

roles/scout.md

@@ -11,6 +11,14 @@ and config.md.
   tier and repository, every reader and writer of the data, the states
   existing records are in. Query real data read-only where a tool and
   permission exist; redact personal data.
+- Changed identifiers: for each name or value the change renames, retires,
+  redirects or reshapes (a route or path, an event, message or log key, an
+  enum value, a column, a config key, an exported symbol), search the literal
+  value as well as the symbol, in every repository, and list each consumer —
+  including those that only compare it: loggers, audit and analytics events,
+  permission checks, watchers, feature flags, links from other apps. One row
+  per consumer: file:line, what it does with the value, what happens to it
+  after the change.
 - Observe rather than infer: run the command, open the screen, call the API.
   For deployed behavior read the deployed revision (`git show
   <deploy-ref>:<path>`), not the working tree.

templates/config.md

@@ -5,7 +5,7 @@ kit: <path to the kit>
 build:
 test:
 lint:
-# start the app for end-to-end scenarios, and the project's own e2e command
+# start the app for end-to-end scenarios (QA runs it detached, outliving the run), and the project's own e2e command
 run:
 e2e:
 # how QA drives the interface: a browser driver, an HTTP client, the shell; empty = whatever the agent has