bloat: BASELINE block fires once per CC session (v1.69.0)

.claude-plugin/marketplace.json

@@ -13,7 +13,7 @@
       "name": "sdlc-wizard",
       "source": ".",
       "description": "SDLC enforcement for AI agents — TDD, planning, self-review, CI shepherd",
-      "version": "1.68.0",
+      "version": "1.69.0",
       "author": {
         "name": "Stefan Ayala"
       },

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "sdlc-wizard",
-  "version": "1.68.0",
+  "version": "1.69.0",
   "description": "SDLC enforcement for AI agents — TDD, planning, self-review, CI shepherd",
   "author": {
     "name": "Stefan Ayala",

.github/workflows/ci.yml

@@ -214,6 +214,9 @@ jobs:
       - name: Run prompt-hook-fires-once tests (#224)
         run: ./tests/test-prompt-hook-fires-once.sh
 
+      - name: Run baseline-fires-once-per-session tests (token bloat audit)
+        run: ./tests/test-baseline-fires-once.sh
+
       - name: Run community scanner tests (#207)
         run: ./tests/test-community-scanner.sh
 

.reviews/baseline-fires-once-001/round-1-review.md

@@ -0,0 +1,34 @@
+**Findings**
+
+1. **P1**: Same-session BASELINE suppression is not concurrency-safe.
+   Evidence: [sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:131) checks for the sentinel before emit, but [sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:164) writes it only after the `cat` block. A 50-process same-session race produced `baseline_outputs=13`, `sentinels=1`.
+   Certify condition: make the first-writer decision atomic, with best-effort fallback preserved, and add a parallel same-`session_id` regression test expecting exactly one BASELINE.
+
+2. **P1**: Valid `session_id` input is ignored when `jq` is unavailable or broken.
+   Evidence: [sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:47) only reads stdin when `jq` exists, and [sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:51) extracts `session_id` through `jq`. With a fake failing `jq` and valid JSON stdin: `first_baseline=1`, `second_baseline=1`, `sentinels=0`.
+   Certify condition: either make `session_id` extraction work without optional `jq`, or explicitly enforce/document `jq` as a runtime dependency and test the dependency behavior.
+
+**Checklist Evidence**
+
+(a) Sequential gate shape is correct: sentinel check at lines 122-132, BASELINE at 137-158, write after emit at 159-168. EFFORT is outside at 90-100; SETUP is outside at 103-112.
+
+(b) SETUP missing test uses sibling tmpdir: [test-baseline-fires-once.sh](/Users/stefanayala/sdlc-wizard/tests/test-baseline-fires-once.sh:110).
+
+(c) Sanitization uses `tr -cd 'A-Za-z0-9._-'` at line 128. Manual malicious id created only `baseline-shown-....badtouchtmpownedstefanayala` inside cache.
+
+(d) Prune is scoped to cache dir and filename pattern: line 167.
+
+(e) Best-effort cache failure verified: cache-dir-is-file manual run returned `rc=0`, `stderr_bytes=0`, `baseline_count=1`.
+
+(f) Version bump grep verified all 7 sites: `package.json:3`, `plugin.json:3`, `marketplace.json:16`, `SDLC.md:1`, `SDLC.md:10`, `CLAUDE_CODE_SDLC_WIZARD.md:2979`, `CLAUDE_CODE_SDLC_WIZARD.md:4058`.
+
+(g) CI path is correct: [.github/workflows/ci.yml](/Users/stefanayala/sdlc-wizard/.github/workflows/ci.yml:218) runs `./tests/test-baseline-fires-once.sh`.
+
+(h) Tests run:
+`./tests/test-baseline-fires-once.sh`: 8 passed, 0 failed.
+`./tests/test-audit-session-load.sh`: 9 passed, 0 failed.
+`./tests/test-hooks.sh`: 154 passed, 0 failed.
+
+Score: **6/10**
+
+**NOT CERTIFIED**

.reviews/baseline-fires-once-001/round-2-review.md

@@ -0,0 +1,16 @@
+No blocking findings.
+
+Finding 1: FIXED. The BASELINE claim is now atomic via noclobber at [hooks/sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:158), with the correct fallback tree: winner emits, existing sentinel suppresses, cache failure emits. The 50-parallel regression at [tests/test-baseline-fires-once.sh](/Users/stefanayala/sdlc-wizard/tests/test-baseline-fires-once.sh:190) passed and counted exactly one `SDLC BASELINE:`.
+
+Finding 2: FIXED. Stdin is read independently of `jq`, `session_id` is extracted with grep/sed at [hooks/sdlc-prompt-check.sh](/Users/stefanayala/sdlc-wizard/hooks/sdlc-prompt-check.sh:51), and `jq` is only used for prompt parsing. The no-`jq` regression at [tests/test-baseline-fires-once.sh](/Users/stefanayala/sdlc-wizard/tests/test-baseline-fires-once.sh:218) passed. I also manually checked an escaped `session_id` mention inside `prompt`; it used the real top-level session id and suppressed the second fire.
+
+Prior passes still hold: EFFORT and SETUP remain outside the BASELINE gate, setup-missing uses a sibling tmpdir, `tr -cd 'A-Za-z0-9._-'` sanitization remains in place, prune is scoped to `baseline-shown-*`, cache-file fallback returned `rc=0`, `baseline=1`, `stderr_bytes=0`, version bumps are consistent at `1.69.0`, and CI runs `./tests/test-baseline-fires-once.sh`.
+
+Verification run:
+`./tests/test-baseline-fires-once.sh`: 10 passed  
+`./tests/test-audit-session-load.sh`: 9 passed  
+`./tests/test-hooks.sh`: 154 passed
+
+Notes for next review: `.reviews/response.json` does not match this handoff; it references `roadmap-96-phase2-001` and a different F-01. I used the handoff’s `fixes_applied` plus the actual previous review findings for this recheck.
+
+Score: 9/10, CERTIFIED

.reviews/preflight-baseline-fires-once-001.md

@@ -0,0 +1,53 @@
+# Preflight Self-Review: BASELINE block fires once per CC session
+
+## What changed
+
+`hooks/sdlc-prompt-check.sh`:
+- Extract `session_id` from stdin JSON alongside existing `prompt` extraction.
+- Gate the static SDLC BASELINE `cat << 'EOF'` block on a per-session sentinel `$SDLC_WIZARD_CACHE_DIR/baseline-shown-<safe_sid>`.
+- After first emit, `mkdir -p` cache dir + `touch` sentinel; prune sentinels older than 7d via `find -mtime +7 -delete`.
+- Sanitize session_id with `tr -cd 'A-Za-z0-9._-'` (defense-in-depth — CC session_ids are UUIDs, but never trust stdin).
+- Fallback when no session_id (legacy CC stdin or direct shell test): emit BASELINE every fire (current behavior).
+
+## What did NOT change
+
+- `_find-sdlc-root.sh` walk-up logic
+- Effort-bump signal detector + nudge (lines 38-99)
+- SETUP-NOT-COMPLETE warning (lines 103-112)
+- ROADMAP #224 SDLC_HOOK_FIRE_LOG instrumentation
+- Plugin-vs-project dedupe heuristic
+- Any other hook in `hooks/`
+
+## Self-review checklist
+
+- [x] `tests/test-baseline-fires-once.sh` — 8/8 PASS (8 cases: first-fire, suppression, different-session re-emit, no-session-id back-compat, SETUP persistence, EFFORT-bump persistence, cache-isolation, byte-shrink)
+- [x] `tests/test-hooks.sh` — 154/154 PASS (no regression in existing hook test suite)
+- [x] `tests/test-prompt-hook-fires-once.sh` — 6/6 PASS (ROADMAP #224 instrumentation regression, includes byte-identical assertion)
+- [x] `tests/test-audit-session-load.sh` — 9/9 PASS (`skills/update/SKILL.md` still under 5K-token threshold)
+- [x] `tests/test-cli.sh` — 78/78 PASS
+- [x] `tests/test-plugin.sh` — 25/25 PASS
+- [x] `tests/test-doc-consistency.sh` — 35/35 PASS
+- [x] Cache writes are best-effort: unwritable cache → falls back to current behavior (BASELINE keeps emitting), never errors to user
+- [x] session_id sanitized before use in filename
+- [x] Comment block on the gating logic explains the *why* (12K token saving, skill duplication once auto-invoked) and the constraints (SETUP/EFFORT-bump must keep firing)
+- [x] Version bumped 1.68.0 → 1.69.0 across 7 metadata sites
+- [x] `CHANGELOG.md` v1.69.0 entry written with behavior matrix
+- [x] `skills/update/SKILL.md` changelog list updated with 1.69.0 entry, older entries collapsed to keep under 5K threshold
+- [x] `.github/workflows/ci.yml` wires new test into validate job
+
+## Specific things to verify in review
+
+1. **Race conditions:** if CC parallelism could fire two `UserPromptSubmit` hooks concurrently for the same session_id, both could check `[ -f $sentinel ]` as false, both emit, both touch. Worst case is BASELINE emits twice on a single rare race. Acceptable, but flag if there's a cleaner pattern.
+
+2. **Stale-cache leakage across CC restarts:** sentinel persists on disk. If a user runs the same CC `session_id` again after a restart (does CC ever reuse session_ids? — CC sessions are UUIDs, almost certainly unique), they'd see no BASELINE on their first prompt. Mitigation: 7-day prune. Verify the worst case is acceptable.
+
+3. **Filename injection / path traversal:** session_id from stdin sanitized via `tr -cd 'A-Za-z0-9._-'`. Confirm this is sufficient. CC session_ids are UUIDs (lowercase hex + dashes), so the strip is purely defense-in-depth.
+
+4. **Byte-shrink test (test 8):** asserts second fire output is `<` (first / 5). Currently 823 → 0 chars. If anyone adds new conditional output that fires post-suppression, the test will catch it. Verify the assertion is the right shape.
+
+5. **Back-compat:** `tests/test-hooks.sh` Test 5 (line ~91 `test_sdlc_hook_size`) calls `"$HOOKS_DIR/sdlc-prompt-check.sh"` with NO stdin. The hook reaches `[ ! -t 0 ]` check, finds tty (when run from terminal/test), skips stdin parse, `SESSION_ID` stays empty, BASELINE emits every fire. Test still passes. Confirm I haven't created a hidden coupling.
+
+## Known limitations
+
+- Sentinel doesn't survive `rm -rf $SDLC_WIZARD_CACHE_DIR`. User who clears cache mid-session sees BASELINE re-emit on next prompt — acceptable.
+- 7-day prune is on-emit, not periodic. A long-idle session whose sentinel ages out then resumes would see one re-emit. Acceptable.

.reviews/response.json

@@ -1,14 +1,19 @@
 {
-  "review_id": "roadmap-96-phase2-001",
-  "round": 3,
-  "responding_to": ".reviews/latest-review.md",
+  "review_id": "baseline-fires-once-001",
+  "round": 2,
+  "responding_to": ".reviews/baseline-fires-once-001/round-1-review.md",
   "responses": [
     {
-      "finding": "F-01",
-      "severity": "P1",
+      "finding": "1",
       "action": "FIXED",
-      "summary": "Round 2 nit closed. Added test_strip_paths_gate_uses_candidate_strip_dir which runs shepherd with --compare-baseline AND --strip-paths, mocks ground-truth.sh to log argv, and asserts both paths are tmpdir-rooted: baseline matches `sdlc-baseline-strip.*tests/e2e/fixtures/test-repo$`, candidate matches `sdlc-candidate-strip.*tests/e2e/fixtures/test-repo$`. Without my fix, the candidate path would be cwd-relative `tests/e2e/fixtures/test-repo` — that mutation would fail this test.",
-      "files": ["tests/test-local-shepherd.sh:test_strip_paths_gate_uses_candidate_strip_dir"]
+      "summary": "Atomic claim via subshell `set -C` (noclobber) + `: > sentinel` redirect. Across N parallel fires with the same session_id, exactly one wins the claim and emits; the rest see file-exists and suppress. New regression test (test_concurrency_same_session_emits_once) fires 50 parallel invocations and asserts exactly 1 BASELINE in aggregated stdout.",
+      "evidence": "hooks/sdlc-prompt-check.sh:131 (claim block), tests/test-baseline-fires-once.sh:166-189 (50-parallel test)"
+    },
+    {
+      "finding": "2",
+      "action": "FIXED",
+      "summary": "Decoupled session_id extraction from jq. Stdin is now read whenever stdin is non-tty (regardless of jq); session_id extracted via grep/sed. Prompt extraction still requires jq (multi-line + arbitrary escape content). New regression test (test_session_id_works_without_jq) builds a restricted PATH excluding jq, fires hook twice with valid session_id JSON stdin, asserts first emits / second suppresses.",
+      "evidence": "hooks/sdlc-prompt-check.sh:46-66 (decoupled extraction), tests/test-baseline-fires-once.sh:191-228 (no-jq regression test using type -P for binary-only resolution)"
     }
   ]
 }

CHANGELOG.md

@@ -4,6 +4,31 @@ All notable changes to the SDLC Wizard.
 
 > **Note:** This changelog is for humans to read. Don't manually apply these changes - just run the wizard ("Check for SDLC wizard updates") and it handles everything automatically.
 
+## [1.69.0] - 2026-05-04
+
+### Token-bloat fix: BASELINE block fires once per CC session
+
+Cuts ~12K tokens/session of duplicate context for users with >3 prompts. The `SDLC BASELINE` block in `hooks/sdlc-prompt-check.sh` (~250 tokens) was firing on every `UserPromptSubmit` — once Claude has the SDLC skill auto-invoked (covers TodoWrite/confidence/workflow phases), every subsequent re-emission is pure duplication. Now gated by a per-`session_id` sentinel under `$SDLC_WIZARD_CACHE_DIR/baseline-shown-<id>`, pruned at 7d.
+
+### Behavior
+
+- **First prompt of a CC session** → BASELINE emits as before (cold-start nudge survives).
+- **Subsequent prompts (same session_id)** → BASELINE suppressed.
+- **New CC session (different session_id)** → BASELINE re-emits.
+- **No session_id in stdin** (legacy CC, direct shell tests) → BASELINE emits every fire (back-compat preserved).
+- `SETUP NOT COMPLETE` warning + `EFFORT BUMP REQUIRED` nudge **continue to fire every prompt** — they're dynamic state warnings, not static reminders.
+
+### Files
+
+- `hooks/sdlc-prompt-check.sh` — extracts `session_id` from stdin JSON; gates the static BASELINE block via per-session sentinel; prunes >7d sentinels on emit.
+- `tests/test-baseline-fires-once.sh` (new — 8 cases covering first-fire, suppression, different-session re-emit, no-session-id back-compat, SETUP-warning persistence, EFFORT-bump persistence, cross-cache-dir isolation, byte-shrink verification).
+- `.github/workflows/ci.yml` — wires new test into `validate` job.
+- `CHANGELOG.md`, `SDLC.md`, `skills/update/SKILL.md`, `package.json`, `.claude-plugin/plugin.json` + `marketplace.json`, `CLAUDE_CODE_SDLC_WIZARD.md` (1.68.0 → 1.69.0).
+
+### Notes
+
+Discovered during ROADMAP #236 functional-bloat audit. Identified `sdlc-prompt-check.sh` as the #1 amplifier (every-prompt × 22 lines × N prompts). Audit method: measure cost × frequency, judge value — not blind delete-and-see. Per-prompt BASELINE failed cost/value once skill is loaded; conditional warnings + effort-bump detector earned their keep and stayed untouched. Other hooks (`model-effort-check`, `precompact-seam-check`, `token-spike-check`) are silent at healthy state — not bloat.
+
 ## [1.68.0] - 2026-05-04
 
 ### Closed (paperwork-stale roadmap rows)

CLAUDE_CODE_SDLC_WIZARD.md

@@ -2976,7 +2976,7 @@ If deployment fails or post-deploy verification catches issues:
 
 **SDLC.md:**
 ```markdown
-<!-- SDLC Wizard Version: 1.68.0 -->
+<!-- SDLC Wizard Version: 1.69.0 -->
 <!-- Setup Date: [DATE] -->
 <!-- Completed Steps: step-0.1, step-0.2, step-0.4, step-1, step-2, step-3, step-4, step-5, step-6, step-7, step-8, step-9 -->
 <!-- Git Workflow: [PRs or Solo] -->
@@ -4055,7 +4055,7 @@ Walk through updates? (y/n)
 Store wizard state in `SDLC.md` as metadata comments (invisible to readers, parseable by Claude):
 
 ```markdown
-<!-- SDLC Wizard Version: 1.68.0 -->
+<!-- SDLC Wizard Version: 1.69.0 -->
 <!-- Setup Date: 2026-01-24 -->
 <!-- Completed Steps: step-0.1, step-0.2, step-1, step-2, step-3, step-4, step-5, step-6, step-7, step-8, step-9 -->
 <!-- Git Workflow: PRs -->

CONTRIBUTING.md

@@ -40,6 +40,7 @@ Thank you for your interest in improving the SDLC Wizard!
    ./tests/test-local-shepherd.sh && \
    ./tests/test-repo-complexity.sh && \
    ./tests/test-prompt-hook-fires-once.sh && \
+   ./tests/test-baseline-fires-once.sh && \
    ./tests/test-community-scanner.sh && \
    ./tests/test-community-fetch.sh && \
    ./tests/test-ground-truth.sh && \

SDLC.md

@@ -1,4 +1,4 @@
-<!-- SDLC Wizard Version: 1.68.0 -->
+<!-- SDLC Wizard Version: 1.69.0 -->
 <!-- Setup Date: 2026-01-24 -->
 <!-- Completed Steps: step-0.1, step-0.2, step-1, step-2, step-3, step-4, step-5, step-6, step-7, step-8, step-9 -->
 # SDLC Configuration
@@ -7,7 +7,7 @@
 
 | Property | Value |
 |----------|-------|
-| Wizard Version | 1.68.0 |
+| Wizard Version | 1.69.0 |
 | Last Updated | 2026-05-04 |
 | Claude Code Baseline | v2.1.111+ (required for Opus 4.7 / `opus[1m]`) |
 | Recommended Model | `opus[1m]` (Opus 4.7, 1M context) — run `/model opus[1m]` |

hooks/sdlc-prompt-check.sh

@@ -43,10 +43,25 @@ fi
 EFFORT_CACHE_DIR="${SDLC_WIZARD_CACHE_DIR:-$HOME/.cache/sdlc-wizard}"
 EFFORT_SIGNALS="$EFFORT_CACHE_DIR/effort-signals.log"
 PROMPT_TEXT=""
-if [ ! -t 0 ] && command -v jq > /dev/null 2>&1; then
+SESSION_ID=""
+# Read stdin once regardless of jq availability — session_id extraction
+# is jq-independent (Codex round 1 P1: BASELINE gate failed when jq was
+# missing or broken). Prompt extraction still needs jq because prompt
+# content can contain arbitrary multi-line text + escapes.
+if [ ! -t 0 ]; then
     STDIN_JSON=$(cat)
     if [ -n "$STDIN_JSON" ]; then
-        PROMPT_TEXT=$(printf '%s' "$STDIN_JSON" | jq -r '.prompt // empty' 2>/dev/null) || PROMPT_TEXT=""
+        # session_id is a UUID-shaped string with no escapable content
+        # in CC's stdin contract — regex extraction is sufficient.
+        # `tr -cd` later strips anything filename-unsafe, so a malformed
+        # input cannot escape the cache dir.
+        SESSION_ID=$(printf '%s' "$STDIN_JSON" \
+            | grep -o '"session_id"[[:space:]]*:[[:space:]]*"[^"]*"' \
+            | head -1 \
+            | sed 's/.*"\([^"]*\)"$/\1/')
+        if command -v jq > /dev/null 2>&1; then
+            PROMPT_TEXT=$(printf '%s' "$STDIN_JSON" | jq -r '.prompt // empty' 2>/dev/null) || PROMPT_TEXT=""
+        fi
     fi
 fi
 if [ -n "$PROMPT_TEXT" ]; then
@@ -109,7 +124,49 @@ SETUP
     exit 0
 fi
 
-cat << 'EOF'
+# Token-bloat fix: BASELINE block fires once per CC session (~250 tok × 50
+# prompts = ~12K wasted tokens before this gate). Once Claude has the SDLC
+# skill auto-invoked (covers TodoWrite/confidence/workflow), this static
+# block is duplicate context. Sentinel is per-session_id so a fresh CC
+# session re-emits the cold-start nudge. Without session_id (legacy CC, or
+# direct shell tests with no JSON stdin), behavior is unchanged — emits
+# every fire. SETUP-not-complete + EFFORT-bump branches above are NOT
+# gated; they're dynamic state warnings that must fire every prompt.
+#
+# Concurrency: claim is atomic via `set -C` (noclobber) — the redirect
+# `: > "$path"` create-or-fails. Across N parallel fires with the same
+# session_id, exactly one wins the claim and emits BASELINE; the rest
+# see file-exists and suppress. (Codex round 1 P1: previous "check then
+# write after emit" pattern allowed N parallel fires to all emit.)
+SHOULD_EMIT_BASELINE=1
+BASELINE_SENTINEL=""
+if [ -n "$SESSION_ID" ]; then
+    BASELINE_CACHE_DIR="${SDLC_WIZARD_CACHE_DIR:-$HOME/.cache/sdlc-wizard}"
+    # Strip path-traversal chars from session_id before using in filename
+    # (defense-in-depth — CC session_ids are UUIDs, but never trust stdin).
+    SAFE_SID=$(printf '%s' "$SESSION_ID" | tr -cd 'A-Za-z0-9._-')
+    if [ -n "$SAFE_SID" ]; then
+        BASELINE_SENTINEL="$BASELINE_CACHE_DIR/baseline-shown-${SAFE_SID}"
+        mkdir -p "$BASELINE_CACHE_DIR" 2>/dev/null || true
+        # Atomic create-or-fail: subshell sets noclobber so `: > "$path"`
+        # fails (rc≠0) if the file already exists. The full conditional
+        # tree:
+        #   - claim succeeds → emit (we won the race)
+        #   - claim fails AND file exists → suppress (someone else won)
+        #   - claim fails AND file doesn't exist → cache unwritable;
+        #     fall back to emit so user never loses cold-start nudge.
+        if (set -C; : > "$BASELINE_SENTINEL") 2>/dev/null; then
+            SHOULD_EMIT_BASELINE=1
+        elif [ -f "$BASELINE_SENTINEL" ]; then
+            SHOULD_EMIT_BASELINE=0
+        else
+            SHOULD_EMIT_BASELINE=1
+        fi
+    fi
+fi
+
+if [ "$SHOULD_EMIT_BASELINE" -eq 1 ]; then
+    cat << 'EOF'
 SDLC BASELINE:
 1. TodoWrite FIRST (plan tasks before coding)
 2. STATE CONFIDENCE: HIGH/MEDIUM/LOW
@@ -130,3 +187,9 @@ Workflow phases:
 
 Quick refs: SDLC.md | TESTING.md | *_PLAN.md for feature
 EOF
+    # Prune sentinels older than 7d so cache doesn't grow forever.
+    # Best-effort: errors silently swallowed.
+    if [ -n "$BASELINE_SENTINEL" ]; then
+        find "$BASELINE_CACHE_DIR" -name 'baseline-shown-*' -type f -mtime +7 -delete 2>/dev/null || true
+    fi
+fi

package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentic-sdlc-wizard",
-  "version": "1.68.0",
+  "version": "1.69.0",
   "description": "SDLC enforcement for Claude Code — hooks, skills, and wizard setup in one command",
   "bin": {
     "sdlc-wizard": "cli/bin/sdlc-wizard.js"