diff --git a/.autoresearch/autoresearch.jsonl b/.autoresearch/autoresearch.jsonl
index af9a7866..e99ad71b 100644
--- a/.autoresearch/autoresearch.jsonl
+++ b/.autoresearch/autoresearch.jsonl
@@ -1,28 +1,25 @@
-{"type":"config","name":"create-pr-optimize","metricName":"total_bytes","metricUnit":"bytes","bestDirection":"lower"}
-{"run":1,"commit":"d018e0f","metric":9073,"metrics":{"line_count":284,"file_count":6,"word_count":1375},"status":"keep","description":"baseline","timestamp":1775130716,"segment":0}
-{"run":2,"commit":"9e5692e","metric":6268,"metrics":{"line_count":183,"file_count":6,"word_count":917},"status":"keep","description":"compress scripts - remove verbose messages and redundant comments","timestamp":1775130811,"segment":0}
-{"run":3,"commit":"17e69a5","metric":5836,"metrics":{"line_count":169,"file_count":6,"word_count":844},"status":"keep","description":"compress SKILL.md prose","timestamp":1775130856,"segment":0}
-{"run":4,"commit":"6265fd9","metric":4534,"metrics":{"line_count":141,"file_count":5,"word_count":666},"status":"keep","description":"remove unused verify-pr-status.sh","timestamp":1775130894,"segment":0}
-{"run":5,"commit":"3eb5c55","metric":4019,"metrics":{"line_count":122,"file_count":4,"word_count":601},"status":"keep","description":"merge sync-with-base into preflight-check","timestamp":1775130966,"segment":0}
-{"run":6,"commit":"d790d50","metric":3558,"metrics":{"line_count":107,"file_count":3,"word_count":541},"status":"keep","description":"inline lib.sh into preflight, remove lib.sh","timestamp":1775131007,"segment":0}
-{"run":7,"commit":"9706617","metric":3202,"metrics":{"line_count":85,"file_count":3,"word_count":490},"status":"keep","description":"further compress SKILL.md","timestamp":1775131034,"segment":0}
-{"run":8,"commit":"9725788","metric":3103,"metrics":{"line_count":82,"file_count":3,"word_count":471},"status":"keep","description":"compact wait-for-merge.sh","timestamp":1775131065,"segment":0}
-{"run":9,"commit":"7f78174","metric":2884,"metrics":{"line_count":68,"file_count":3,"word_count":448},"status":"keep","description":"further compress preflight-check.sh","timestamp":1775131088,"segment":0}
-{"type":"config","name":"create-pr-skill-tokens","metricName":"skill_bytes","metricUnit":"bytes","bestDirection":"lower"}
-{"run":10,"commit":"1b650ac","metric":1081,"metrics":{"skill_lines":27,"skill_words":151,"script_bytes":1803},"status":"keep","description":"baseline (segment 1 — skill_bytes only)","timestamp":1775131168,"segment":1}
-{"run":11,"commit":"6ba0c3d","metric":802,"metrics":{"skill_lines":23,"skill_words":110,"script_bytes":1803},"status":"keep","description":"compress SKILL.md — remove redundant sections","timestamp":1775131195,"segment":1}
-{"run":12,"commit":"ec416bc","metric":732,"metrics":{"skill_lines":19,"skill_words":100,"script_bytes":1803},"status":"keep","description":"extract script path variable S=","timestamp":1775131213,"segment":1}
-{"run":13,"commit":"9bb6f1e","metric":675,"metrics":{"skill_lines":18,"skill_words":93,"script_bytes":1803},"status":"keep","description":"merge inline comments, remove bold","timestamp":1775131235,"segment":1}
-{"run":14,"commit":"563874d","metric":635,"metrics":{"skill_lines":18,"skill_words":82,"script_bytes":1803},"status":"keep","description":"micro-compress wording","timestamp":1775131249,"segment":1}
-{"run":15,"commit":"059de59","metric":605,"metrics":{"skill_lines":17,"skill_words":82,"script_bytes":1803},"status":"keep","description":"remove explicit template path","timestamp":1775131290,"segment":1}
-{"run":16,"commit":"059de59","metric":608,"metrics":{"skill_lines":16,"skill_words":83,"script_bytes":1803},"status":"discard","description":"merge pr create+merge into one line (bytes increased)","timestamp":1775131309,"segment":1}
-{"run":17,"commit":"96b1a8f","metric":665,"metrics":{"skill_lines":17,"skill_words":92,"script_bytes":1818},"status":"keep","description":"add auto-merge re-enable after CI fix + push -u","timestamp":1775132341,"segment":1}
-{"run":18,"commit":"17c7ef7","metric":665,"metrics":{"skill_lines":17,"skill_words":92,"script_bytes":1818},"status":"keep","description":"edge test: main branch — agent skipped script, did manual logic","timestamp":1775132900,"segment":1}
-{"run":19,"commit":"b2e1f87","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1818},"status":"keep","description":"clarify scripts MUST be run (test-driven fix)","timestamp":1775132948,"segment":1}
-{"run":20,"commit":"b2e1f87","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1818},"status":"keep","description":"edge test: nothing-to-commit — agent stopped correctly but preflight ran unnecessarily","timestamp":1775133044,"segment":1}
-{"run":22,"commit":"b2e1f87","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1818},"status":"keep","description":"must-run test passed, found broken tests referencing deleted scripts","timestamp":1775133359,"segment":1}
-{"run":23,"commit":"dece665","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1818},"status":"keep","description":"fix broken tests for deleted scripts — 63/63 pass","timestamp":1775133491,"segment":1}
-{"run":25,"commit":"dece665","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1818},"status":"keep","description":"final validation — both scripts executed, 14 tool calls, PR #607 merged","timestamp":1775133750,"segment":1}
-{"run":26,"commit":"6a0be38","metric":794,"metrics":{"skill_lines":20,"skill_words":109,"script_bytes":1802},"status":"keep","description":"remove --delete-branch from fallback merge","timestamp":1775133769,"segment":1}
-{"run":27,"commit":"494dc0b","metric":776,"metrics":{"skill_lines":20,"skill_words":107,"script_bytes":1802},"status":"keep","description":"remove re-run preflight, reorder main check","timestamp":1775133796,"segment":1}
-{"run":28,"commit":"b792e67","metric":776,"metrics":{"skill_lines":21,"skill_words":107,"script_bytes":1802},"status":"keep","description":"wrap CI fail line for markdownlint","timestamp":1775133928,"segment":1}
+{"type":"config","name":"ralph-optimize","metricName":"skill_bytes","metricUnit":"bytes","bestDirection":"lower"}
+{"run":1,"commit":"c8c2fa1","metric":2914,"metrics":{"skill_lines":90,"cancel_bytes":696,"hook_bytes":3747,"hook_lines":138,"total_bytes":7357},"status":"keep","description":"baseline","timestamp":1775210437,"segment":0}
+{"run":2,"commit":"6934219","metric":1710,"metrics":{"skill_lines":52,"cancel_bytes":696,"hook_bytes":3747,"hook_lines":138,"total_bytes":6153},"status":"keep","description":"compress SKILL.md — flatten JSON, remove redundant prose","timestamp":1775210484,"segment":0}
+{"run":3,"commit":"afeef81","metric":1364,"metrics":{"skill_lines":41,"cancel_bytes":696,"hook_bytes":3747,"hook_lines":138,"total_bytes":5807},"status":"keep","description":"further compress SKILL.md — shorten sections, tighten wording","timestamp":1775210509,"segment":0}
+{"run":4,"commit":"c4b4aba","metric":1364,"metrics":{"skill_lines":41,"cancel_bytes":431,"hook_bytes":3747,"hook_lines":138,"total_bytes":5542},"status":"keep","description":"compress ralph-cancel SKILL.md","timestamp":1775210530,"segment":0}
+{"run":5,"commit":"d70b5f9","metric":1364,"metrics":{"skill_lines":41,"cancel_bytes":431,"hook_bytes":2725,"hook_lines":74,"total_bytes":4520},"status":"keep","description":"simplify ralph-persist.ts — remove interfaces, inline functions","timestamp":1775210565,"segment":0}
+{"run":6,"commit":"01d96d4","metric":1260,"metrics":{"skill_lines":40,"cancel_bytes":431,"hook_bytes":2725,"hook_lines":74,"total_bytes":4416},"status":"keep","description":"micro-compress SKILL.md — merge steps, shorten JSON","timestamp":1775210587,"segment":0}
+{"run":7,"commit":"974f67c","metric":1260,"metrics":{"skill_lines":40,"cancel_bytes":431,"hook_bytes":2157,"hook_lines":52,"total_bytes":3848},"status":"keep","description":"compress ralph-persist.ts — extract helpers, shorten vars","timestamp":1775210616,"segment":0}
+{"run":8,"commit":"0405ab8","metric":1237,"metrics":{"skill_lines":36,"cancel_bytes":431,"hook_bytes":2157,"hook_lines":52,"total_bytes":3825},"status":"keep","description":"replace JSON code block with inline field description","timestamp":1775210641,"segment":0}
+{"run":9,"commit":"465de49","metric":1197,"metrics":{"skill_lines":34,"cancel_bytes":431,"hook_bytes":2157,"hook_lines":52,"total_bytes":3785},"status":"keep","description":"extract $S path var, merge verification+done sections","timestamp":1775210664,"segment":0}
+{"run":10,"commit":"570f903","metric":1193,"metrics":{"skill_lines":36,"cancel_bytes":431,"hook_bytes":2157,"hook_lines":52,"total_bytes":3781},"status":"keep","description":"consolidate flags into single line at bottom","timestamp":1775210689,"segment":0}
+{"run":11,"commit":"fe1ab49","metric":1193,"metrics":{"skill_lines":36,"cancel_bytes":431,"hook_bytes":1962,"hook_lines":48,"total_bytes":3586},"status":"keep","description":"shorten block message, compress variable names","timestamp":1775210722,"segment":0}
+{"run":13,"commit":"823b62e","metric":1249,"metrics":{"skill_lines":32,"cancel_bytes":431,"hook_bytes":1962,"hook_lines":48,"total_bytes":3642},"status":"keep","description":"clarify SKILL.md based on agent test — progress.txt, deslop, review criteria (+123 bytes for clarity)","timestamp":1775211110,"segment":0}
+{"run":14,"commit":"97d4d0b","metric":1272,"metrics":{"skill_lines":31,"cancel_bytes":431,"hook_bytes":1962,"hook_lines":48,"total_bytes":3665},"status":"keep","description":"clarify progress.txt timing, remove redundant parse step","timestamp":1775211256,"segment":0}
+{"run":15,"commit":"59d1f66","metric":1359,"metrics":{"skill_lines":31,"cancel_bytes":431,"hook_bytes":1962,"hook_lines":48,"total_bytes":3752},"status":"keep","description":"define flag behaviors — --no-prd creates minimal prd.json (agent test fix)","timestamp":1775211369,"segment":0}
+{"run":16,"commit":"4ea8184","metric":1396,"metrics":{"skill_lines":31,"cancel_bytes":431,"hook_bytes":1962,"hook_lines":48,"total_bytes":3789},"status":"keep","description":"clarify TDD with existing tests, fix --no-prd wording (agent test fix)","timestamp":1775211475,"segment":0}
+{"run":17,"commit":"0201a7a","metric":1396,"metrics":{"skill_lines":31,"cancel_bytes":525,"hook_bytes":1962,"hook_lines":48,"total_bytes":3883},"status":"keep","description":"add CWD context and hook explanation to ralph-cancel (agent test fix)","timestamp":1775211579,"segment":0}
+{"run":18,"commit":"e6ea5dd","metric":1438,"metrics":{"skill_lines":32,"cancel_bytes":525,"hook_bytes":1962,"hook_lines":48,"total_bytes":3925},"status":"keep","description":"add INSIGHT format to progress.txt, clarify per-story update (agent test)","timestamp":1775211830,"segment":0}
+{"run":19,"commit":"fb6ca04","metric":1438,"metrics":{"skill_lines":32,"cancel_bytes":525,"hook_bytes":2286,"hook_lines":55,"total_bytes":4249},"status":"keep","description":"add story progress to block message (N/M stories done)","timestamp":1775211874,"segment":0}
+{"run":20,"commit":"60ca33c","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2286,"hook_lines":55,"total_bytes":3909},"status":"keep","description":"simplify for universal use — remove $S, add one-at-a-time + slop def. Agent: 7/10","timestamp":1775212030,"segment":0}
+{"run":21,"commit":"52ff4aa","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2286,"hook_lines":55,"total_bytes":3909},"status":"keep","description":"e2e test: /ralph actual invocation — 3 stories, 1 iteration, all pass","timestamp":1775212340,"segment":0}
+{"run":22,"commit":"52ff4aa","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2286,"hook_lines":55,"total_bytes":3909},"status":"keep","description":"e2e test: /ralph --no-prd — single story auto-generated, no confusion, all pass","timestamp":1775212411,"segment":0}
+{"run":23,"commit":"52ff4aa","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2286,"hook_lines":55,"total_bytes":3909},"status":"keep","description":"e2e test: /ralph-cancel — active loop cancelled cleanly, no issues","timestamp":1775212452,"segment":0}
+{"run":24,"commit":"2af754f","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2487,"hook_lines":60,"total_bytes":4110},"status":"keep","description":"include last progress.txt failure in block message","timestamp":1775212490,"segment":0}
+{"run":25,"commit":"72eb631","metric":1098,"metrics":{"skill_lines":29,"cancel_bytes":525,"hook_bytes":2390,"hook_lines":58,"total_bytes":4013},"status":"keep","description":"use Bun.stdin.text() for simpler stdin reading","timestamp":1775212517,"segment":0}
diff --git a/.autoresearch/autoresearch.md b/.autoresearch/autoresearch.md
index b505d00a..284057d5 100644
--- a/.autoresearch/autoresearch.md
+++ b/.autoresearch/autoresearch.md
@@ -1,47 +1,36 @@
-# Autoresearch: create-pr token efficiency
+# Autoresearch: Ralph Plugin Optimization
 
 ## Objective
-Optimize the `plugins/me/skills/create-pr/` skill for token efficiency and correctness. SKILL.md is loaded into LLM context when invoked — fewer bytes = less cost. Scripts run at execution time and don't affect token cost, but must be correct.
+Optimize the `plugins/ralph/` plugin for simplicity and token efficiency. SKILL.md files are loaded into LLM context when invoked — fewer bytes = less cost. The hook (ralph-persist.ts) runs at OS level but should also be minimal and clean.
 
 ## Metrics
-- **Primary**: skill_bytes (bytes, lower is better) — SKILL.md byte count
-- **Secondary**: skill_lines, skill_words, script_bytes
+- **Primary**: skill_bytes (bytes, lower is better) — ralph/SKILL.md byte count
+- **Secondary**: skill_lines, cancel_bytes, hook_bytes, hook_lines, total_bytes
 
 ## How to Run
-`./.autoresearch/run.sh` — outputs `METRIC name=number` lines.
+`./.autoresearch/run.sh` — outputs `METRIC name=number` lines. Validates frontmatter, hooks.json, TS compilation, and BATS tests.
 
 ## Files in Scope
 | File | Purpose |
 |------|---------|
-| `plugins/me/skills/create-pr/SKILL.md` | Main skill definition (loaded into LLM context) |
-| `plugins/me/skills/create-pr/scripts/preflight-check.sh` | Pre-push checks + auto-sync |
-| `plugins/me/skills/create-pr/scripts/wait-for-merge.sh` | Wait for CI + merge |
+| `plugins/ralph/skills/ralph/SKILL.md` | Main skill (LLM context — primary target) |
+| `plugins/ralph/skills/ralph-cancel/SKILL.md` | Cancel skill (LLM context) |
+| `plugins/ralph/hooks/ralph-persist.ts` | Stop hook engine (Bun runtime) |
+| `plugins/ralph/hooks/hooks.json` | Hook registration config |
 
 ## Off Limits
-- Do not break the PR workflow
-- Exit codes must be preserved
+- Do not break the persistence loop workflow (activate → iterate → complete)
+- Cancel signal mechanism must work
+- Session isolation must be preserved
+- Stale state recovery must work
+- Always exit 0 (never crash Claude)
+- plugin.json metadata
 
 ## Constraints
-- Scripts must pass shellcheck
+- Tests must pass (ralph_persist.bats + ralph_hooks_json.bats)
 - SKILL.md must have valid frontmatter
-- Tests must pass (63/63)
+- hooks.json must be valid JSON
+- TypeScript must compile with bun
 
 ## What's Been Tried
-### Structural changes (big wins)
-- Removed unused verify-pr-status.sh (-1302 bytes)
-- Merged sync-with-base.sh into preflight-check.sh (-515 bytes)
-- Inlined lib.sh into preflight-check.sh (-461 bytes)
-
-### SKILL.md compression (medium wins)
-- Removed Overview, When to Use, Stop Conditions sections
-- Extracted S= path variable for script paths
-- Removed bold markdown markers, flattened sections
-
-### Test-driven fixes (increased bytes for correctness)
-- "scripts MUST be run" directive (+129 bytes) — agents were skipping scripts
-- auto-merge re-enable after CI fix (+60 bytes) — tested on PR #604
-- push -u in preflight — new branches had no upstream
-
-### Dead ends
-- Merging gh pr create + merge into one line — bytes increased
-- Further compression below ~700 bytes — losing essential information
+(Starting fresh — no experiments yet)
diff --git a/.autoresearch/dashboard.md b/.autoresearch/dashboard.md
index f59dae12..9bd6e40d 100644
--- a/.autoresearch/dashboard.md
+++ b/.autoresearch/dashboard.md
@@ -1,31 +1,29 @@
-# Autoresearch Dashboard: create-pr-optimize
+# Autoresearch Dashboard: ralph-optimize
 
-## Segment 1: skill_bytes (SKILL.md only)
-**Runs:** 14 | **Kept:** 12 | **Discarded:** 1 | **Tests:** 1
-**Baseline:** 1081 bytes (#10)
-**Best pure:** 605 bytes (#15, -44.0%)
-**Current:** 794 bytes (#19, -26.5%) — includes test-driven fixes
+**Runs:** 20 | **Kept:** 17 | **Discarded:** 0 | **Crashed:** 0
+**Baseline:** skill=2914, total=7357 bytes
+**Final:** skill=1098 (-62.3%), total=3909 (-46.9%)
+**Agent clarity:** 7/10
 
-| # | commit | skill_bytes | status | description |
-|---|--------|-------------|--------|-------------|
-| 10 | 1b650ac | 1081 | keep | baseline |
-| 11 | 6ba0c3d | 802 (-25.8%) | keep | remove redundant sections |
-| 12 | ec416bc | 732 (-32.3%) | keep | extract S= path variable |
-| 13 | 9bb6f1e | 675 (-37.6%) | keep | merge comments, remove bold |
-| 14 | 563874d | 635 (-41.3%) | keep | micro-compress wording |
-| 15 | 059de59 | 605 (-44.0%) | keep | remove template path |
-| 16 | 059de59 | 608 | discard | merge create+merge lines |
-| 17 | 96b1a8f | 665 | keep | add auto-merge re-enable (test fix) |
-| 18 | - | - | test | edge: main branch — agent skipped scripts |
-| 19 | b2e1f87 | 794 | keep | "scripts MUST be run" directive |
-| 20 | - | - | test | edge: nothing-to-commit — agent handled correctly |
-| 22 | - | - | test | must-run directive confirmed working |
-| 23 | dece665 | 794 | keep | fix broken tests — 63/63 pass |
+| # | skill | cancel | hook | total | description |
+|---|-------|--------|------|-------|-------------|
+| 1 | 2914 | 696 | 3747 | 7357 | baseline |
+| 2 | 1710 | 696 | 3747 | 6153 | compress SKILL.md |
+| 3 | 1364 | 696 | 3747 | 5807 | further compress |
+| 5 | 1364 | 431 | 2725 | 4520 | simplify hook |
+| 7 | 1260 | 431 | 2157 | 3848 | compress hook |
+| 11 | 1193 | 431 | 1962 | 3586 | best pure compression |
+| 15 | 1359 | 431 | 1962 | 3752 | +clarity (agent fixes) |
+| 19 | 1438 | 525 | 2286 | 4249 | +story progress in hook |
+| 20 | 1098 | 525 | 2286 | 3909 | final: simplified for universal use |
 
-## Subagent Test Results
-| PR | Scenario | Result | Finding |
-|----|----------|--------|---------|
-| #601-602 | basic flow (main SKILL) | pass | - |
-| #604 | optimized SKILL | pass | auto-merge disabled after push, push -u needed |
-| #605 | main branch edge | pass | agent skipped scripts (fixed with MUST directive) |
-| #606 | MUST directive test | pass | scripts executed correctly, CI failed on stale tests |
+## Agent Tests (7 total, all pass)
+| Scenario | Clarity Finding |
+|----------|----------------|
+| Simple task | progress.txt init, deslop vague |
+| Multi-story | one-at-a-time unclear |
+| --no-prd | "skip PRD" contradictory |
+| Edge cases | retry flow works |
+| ralph-cancel | CWD + hook explanation needed |
+| Calculator | 7/10 — clean |
+| Calculator --no-prd | 7/10 — auto-generate rule wanted |
diff --git a/.autoresearch/run.sh b/.autoresearch/run.sh
index 32c9c6a7..e0ecdcae 100755
--- a/.autoresearch/run.sh
+++ b/.autoresearch/run.sh
@@ -1,33 +1,43 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-SKILL="plugins/me/skills/create-pr/SKILL.md"
-SCRIPTS_DIR="plugins/me/skills/create-pr/scripts"
+ROOT="plugins/ralph"
 
-# Primary: SKILL.md bytes (this is what loads into LLM context)
-SKILL_BYTES=$(wc -c < "$SKILL" | tr -d ' ')
-echo "METRIC skill_bytes=$SKILL_BYTES"
+# Pre-check: files exist
+for f in "$ROOT/skills/ralph/SKILL.md" "$ROOT/skills/ralph-cancel/SKILL.md" "$ROOT/hooks/ralph-persist.ts" "$ROOT/hooks/hooks.json" "$ROOT/.claude-plugin/plugin.json"; do
+  [[ -f "$f" ]] || { echo "MISSING: $f" >&2; exit 1; }
+done
+
+# Skill bytes (loaded into LLM context — primary optimization target)
+SKILL_BYTES=$(wc -c < "$ROOT/skills/ralph/SKILL.md" | tr -d ' ')
+SKILL_LINES=$(wc -l < "$ROOT/skills/ralph/SKILL.md" | tr -d ' ')
+
+# Cancel skill bytes
+CANCEL_BYTES=$(wc -c < "$ROOT/skills/ralph-cancel/SKILL.md" | tr -d ' ')
+
+# Hook bytes (runtime, not LLM context)
+HOOK_BYTES=$(wc -c < "$ROOT/hooks/ralph-persist.ts" | tr -d ' ')
+HOOK_LINES=$(wc -l < "$ROOT/hooks/ralph-persist.ts" | tr -d ' ')
+
+# Total plugin bytes
+TOTAL=$((SKILL_BYTES + CANCEL_BYTES + HOOK_BYTES))
 
-# Secondary
-SKILL_LINES=$(wc -l < "$SKILL" | tr -d ' ')
+# Validate: SKILL.md has frontmatter
+head -1 "$ROOT/skills/ralph/SKILL.md" | grep -q '^---' || { echo "ERROR: SKILL.md missing frontmatter" >&2; exit 1; }
+
+# Validate: hooks.json is valid JSON
+jq empty "$ROOT/hooks/hooks.json" 2>/dev/null || { echo "ERROR: hooks.json invalid" >&2; exit 1; }
+
+# Validate: TypeScript compiles
+# Validate: TypeScript parses (dry-run with empty stdin)
+echo '{}' | bun run "$ROOT/hooks/ralph-persist.ts" >/dev/null 2>&1 || { echo "ERROR: TS runtime error" >&2; exit 1; }
+
+# Run tests
+bats tests/ralph_persist.bats tests/ralph_hooks_json.bats >/dev/null 2>&1 || { echo "ERROR: tests failed" >&2; exit 1; }
+
+echo "METRIC skill_bytes=$SKILL_BYTES"
 echo "METRIC skill_lines=$SKILL_LINES"
-SKILL_WORDS=$(wc -w < "$SKILL" | tr -d ' ')
-echo "METRIC skill_words=$SKILL_WORDS"
-SCRIPT_BYTES=$(cat "$SCRIPTS_DIR"/*.sh 2>/dev/null | wc -c | tr -d ' ')
-echo "METRIC script_bytes=$SCRIPT_BYTES"
-
-# Validity
-echo "--- Validity Checks ---"
-head -1 "$SKILL" | grep -q '^---' || { echo "FAIL: missing frontmatter" >&2; exit 1; }
-echo "OK: frontmatter"
-
-FAIL=0
-for f in "$SCRIPTS_DIR"/*.sh; do
-  if ! (cd "$SCRIPTS_DIR" && shellcheck -x "$(basename "$f")") >/dev/null 2>&1; then
-    echo "FAIL: shellcheck $(basename "$f")" >&2
-    (cd "$SCRIPTS_DIR" && shellcheck -x "$(basename "$f")") >&2 || true
-    FAIL=1
-  fi
-done
-[[ $FAIL -eq 0 ]] && echo "OK: shellcheck" || exit 1
-echo "--- Done ---"
+echo "METRIC cancel_bytes=$CANCEL_BYTES"
+echo "METRIC hook_bytes=$HOOK_BYTES"
+echo "METRIC hook_lines=$HOOK_LINES"
+echo "METRIC total_bytes=$TOTAL"
diff --git a/.autoresearch/worklog.md b/.autoresearch/worklog.md
index ec63cf69..80d8dbc3 100644
--- a/.autoresearch/worklog.md
+++ b/.autoresearch/worklog.md
@@ -1,47 +1,40 @@
-# Worklog: create-pr token efficiency
+# Worklog: Ralph Plugin Optimization
 
 ## Session Info
-- Started: 2026-04-02
-- Goal: Reduce token cost of create-pr skill while keeping it simple, correct, and effective
-
----
-
-### Segment 0 (total_bytes): Runs 1-9
-Compressed all files from 9073→2884 bytes (-68.2%):
-- Removed verbose error messages/comments in scripts
-- Removed unused verify-pr-status.sh
-- Merged sync-with-base.sh into preflight-check.sh
-- Inlined lib.sh (only used by 1 script)
-
-### Segment 1 (skill_bytes): Runs 10-27
-Re-focused on SKILL.md only (what LLM reads). 1081→776 bytes (-28.2%):
-- Removed redundant sections, shortened description
-- Extracted `S=` path variable
-- Added "scripts MUST be run" directive (test-driven)
-- Added auto-merge re-enable after CI fix (test-driven)
-- Removed redundant "re-run preflight" instruction
-- Fixed broken tests (63/63 pass)
-
-### Subagent Tests (4 PRs)
-| PR | Scenario | Finding |
-|----|----------|---------|
-| #604 | basic optimized flow | push -u needed, auto-merge disabled after push |
-| #605 | main branch | agent skipped scripts → added MUST directive |
-| #606 | MUST directive | scripts executed correctly, stale tests found |
-| #607 | final validation | clean pass, 14 tool calls |
-
-### Bug Fixes Found Through Testing
-1. preflight push needs `-u` for new branches
-2. auto-merge disabled after force-push → added re-enable instruction
-3. agent skipping scripts → added "MUST be run" directive
-4. stale tests referencing deleted scripts → updated test suite
-5. `--delete-branch` in fallback merge inconsistent → removed
-
----
+- **Goal:** Optimize Ralph plugin for simplicity, token efficiency, and clarity
+- **Branch:** autoresearch/ralph-improve-20260403
+- **Started:** 2026-04-03
+- **Primary Metric:** skill_bytes (lower is better)
+
+## Final Results
+- Baseline: skill=2914, cancel=696, hook=3747, total=7357 bytes
+- Final: skill=1098, cancel=525, hook=2286, total=3909 bytes
+- **Net: -46.9% total** (-62.3% skill, -24.6% cancel, -39.0% hook)
+- Agent clarity score: 7/10
+
+## Phase 1: Byte Compression (runs 1-11)
+- Flattened JSON examples, removed redundant sections
+- Simplified TypeScript (removed interfaces, extracted helpers)
+- Shortened variable names, block messages
+- Best pure compression: skill=1193, total=3586
+
+## Phase 2: Agent Testing (runs 13-18)
+7 subagent tests found and fixed:
+1. progress.txt init → "if it exists"
+2. --no-prd undefined → "auto-generate single-story prd"
+3. Deslop vague → "unnecessary comments, dead code, over-abstractions"
+4. One-at-a-time unclear → explicit
+5. ralph-cancel CWD context → added
+6. cancel-signal purpose → hook explanation added
+
+## Phase 3: Simplification (runs 19-20)
+- Removed $S variable, [INSIGHT] format, verbose explanations
+- Added story progress to hook block message (N/M stories done)
+- Final version: clean, minimal, universally applicable
 
 ## Key Insights
-- SKILL.md is the only file that costs tokens — scripts don't load into context
-- "MUST run" directive is essential — without it agents reimplement script logic
-- Real testing (subagent PRs) found 5 bugs that static analysis missed
-- Byte reduction has diminishing returns below ~700 bytes for this skill
-- Code block format is the primary instruction channel for LLM agents
+- Byte compression alone misleads — unclear instructions cost more overall
+- Agent testing catches issues BATS unit tests miss
+- 7/10 clarity is good for a 29-line SKILL.md
+- Remaining 3 points want examples, which conflicts with brevity
+- Hook's story progress (N/M done) gives agents useful context at zero SKILL.md cost
diff --git a/plugins/ralph/hooks/ralph-persist.ts b/plugins/ralph/hooks/ralph-persist.ts
index df71e4da..8cf006b6 100644
--- a/plugins/ralph/hooks/ralph-persist.ts
+++ b/plugins/ralph/hooks/ralph-persist.ts
@@ -2,137 +2,57 @@
 import { existsSync, readFileSync, writeFileSync, unlinkSync, mkdirSync } from "fs";
 import { join } from "path";
 
-const STALE_THRESHOLD_MS = 2 * 60 * 60 * 1000; // 2 hours
+const STALE_MS = 7_200_000; // 2h
 
-interface HookInput {
-  session_id?: string;
-  cwd?: string;
-  hook_event_name?: string;
-}
-
-interface RalphState {
-  active: boolean;
-  session_id: string;
-  iteration: number;
-  max_iterations: number;
-  started_at: string;
-  last_checked_at: string;
-  prompt: string;
-}
-
-function readState(stateDir: string): RalphState | null {
-  const path = join(stateDir, "ralph-state.json");
-  if (!existsSync(path)) return null;
+function allow(): never { process.exit(0); }
+function block(i: number, m: number, p: string, cwd: string): never {
+  let progress = "";
   try {
-    return JSON.parse(readFileSync(path, "utf8")) as RalphState;
-  } catch {
-    return null;
-  }
-}
-
-function writeState(stateDir: string, state: RalphState): void {
-  mkdirSync(stateDir, { recursive: true });
-  writeFileSync(join(stateDir, "ralph-state.json"), JSON.stringify(state, null, 2));
-}
-
-function allow(): void {
-  // Write nothing, exit 0
-  process.exit(0);
-}
-
-function block(iteration: number, maxIterations: number, prompt: string): void {
-  process.stdout.write(
-    JSON.stringify({
-      decision: "block",
-      reason: `[RALPH LOOP - ITERATION ${iteration}/${maxIterations}] Work is not done. Continue working on the task: ${prompt}`,
-    })
-  );
+    const prd = JSON.parse(readFileSync(join(cwd, ".ralph", "prd.json"), "utf8"));
+    const stories = prd.userStories || [];
+    const done = stories.filter((s: any) => s.passes).length;
+    progress = ` (${done}/${stories.length} stories done)`;
+  } catch {}
+  let lastFail = "";
+  try {
+    const lines = readFileSync(join(cwd, ".ralph", "progress.txt"), "utf8").trim().split("\n");
+    lastFail = ` Last: ${lines[lines.length - 1]}`;
+  } catch {}
+  process.stdout.write(JSON.stringify({ decision: "block", reason: `[RALPH ${i}/${m}]${progress}${lastFail} Continue: ${p}` }));
   process.exit(0);
 }
 
-async function main(): Promise<void> {
-  const chunks: Buffer[] = [];
-  for await (const chunk of process.stdin) {
-    chunks.push(chunk as Buffer);
-  }
-  const input: HookInput = JSON.parse(Buffer.concat(chunks).toString("utf8") || "{}");
-
+async function main() {
+  const input = JSON.parse(await Bun.stdin.text() || "{}");
   const cwd = input.cwd ?? process.cwd();
-  const sessionId = input.session_id ?? "";
-  const stateDir = join(cwd, ".ralph", "state");
-
-  // Check activation sentinel
-  const activatingSentinel = join(stateDir, "ralph-activating");
-  if (existsSync(activatingSentinel) && !existsSync(join(stateDir, "ralph-state.json"))) {
-    const prompt = readFileSync(activatingSentinel, "utf8").trim();
-    unlinkSync(activatingSentinel);
+  const sid = input.session_id ?? "";
+  const dir = join(cwd, ".ralph", "state");
+  const sp = join(dir, "ralph-state.json");
+  const save = (s: any) => writeFileSync(sp, JSON.stringify(s, null, 2));
+  const off = (s: any) => { s.active = false; save(s); allow(); };
+
+  const sentinel = join(dir, "ralph-activating");
+  if (existsSync(sentinel) && !existsSync(sp)) {
+    const p = readFileSync(sentinel, "utf8").trim();
+    unlinkSync(sentinel);
     const now = new Date().toISOString();
-    const state: RalphState = {
-      active: true,
-      session_id: sessionId,
-      iteration: 1,
-      max_iterations: 10,
-      started_at: now,
-      last_checked_at: now,
-      prompt,
-    };
-    writeState(stateDir, state);
-    block(1, state.max_iterations, state.prompt);
-    return;
-  }
-
-  const state = readState(stateDir);
-
-  // Pass-through: no state
-  if (!state) {
-    allow();
-    return;
-  }
-
-  // Pass-through: inactive
-  if (!state.active) {
-    allow();
-    return;
-  }
-
-  // Pass-through: session mismatch
-  // If either session_id is empty/unknown, skip the check — don't break a running loop
-  // just because the caller didn't provide a session_id.
-  if (state.session_id && sessionId && state.session_id !== sessionId) {
-    allow();
-    return;
-  }
-
-  // Pass-through: cancel signal
-  const cancelSignal = join(stateDir, "cancel-signal-state.json");
-  if (existsSync(cancelSignal)) {
-    unlinkSync(cancelSignal);
-    state.active = false;
-    writeState(stateDir, state);
-    allow();
-    return;
-  }
-
-  // Pass-through: stale state
-  const lastChecked = new Date(state.last_checked_at).getTime();
-  if (Date.now() - lastChecked > STALE_THRESHOLD_MS) {
-    state.active = false;
-    writeState(stateDir, state);
-    allow();
-    return;
-  }
-
-  // Block: extend max_iterations if needed, then block
-  if (state.iteration >= state.max_iterations) {
-    state.max_iterations += 10;
-  }
-  state.iteration += 1;
-  state.last_checked_at = new Date().toISOString();
-  writeState(stateDir, state);
-  block(state.iteration, state.max_iterations, state.prompt);
+    mkdirSync(dir, { recursive: true });
+    save({ active: true, session_id: sid, iteration: 1, max_iterations: 10, started_at: now, last_checked_at: now, prompt: p });
+    block(1, 10, p, cwd);
+  }
+
+  let s: any;
+  try { s = JSON.parse(readFileSync(sp, "utf8")); } catch { allow(); }
+  if (!s?.active) allow();
+  if (s.session_id && sid && s.session_id !== sid) allow();
+  if (existsSync(join(dir, "cancel-signal-state.json"))) { unlinkSync(join(dir, "cancel-signal-state.json")); off(s); }
+  if (Date.now() - new Date(s.last_checked_at).getTime() > STALE_MS) off(s);
+
+  if (s.iteration >= s.max_iterations) s.max_iterations += 10;
+  s.iteration += 1;
+  s.last_checked_at = new Date().toISOString();
+  save(s);
+  block(s.iteration, s.max_iterations, s.prompt, cwd);
 }
 
-main().catch((err) => {
-  process.stderr.write(`ralph-persist error: ${err}\n`);
-  process.exit(0); // Always exit 0 — never crash Claude
-});
+main().catch((e) => { process.stderr.write(`ralph-persist: ${e}\n`); process.exit(0); });
diff --git a/plugins/ralph/skills/ralph-cancel/SKILL.md b/plugins/ralph/skills/ralph-cancel/SKILL.md
index fbb1ae41..867fa8ca 100644
--- a/plugins/ralph/skills/ralph-cancel/SKILL.md
+++ b/plugins/ralph/skills/ralph-cancel/SKILL.md
@@ -5,19 +5,10 @@ description: Cancel an active Ralph loop — cleans up state files and exits the
 
 # Ralph Cancel
 
-Cancel the active Ralph loop for the current project.
+Cancel the Ralph loop in the current project (CWD).
 
-## Steps
-
-1. Check if `.ralph/state/ralph-state.json` exists
-   - If it does not exist, report: "No active Ralph loop found."
-   - If it exists, read it to confirm `active: true`
-   - If `active: false`, report: "Ralph loop is already inactive."
-
-2. Write `.ralph/state/cancel-signal-state.json` with content `{}`
-
-3. Update `.ralph/state/ralph-state.json`: set `active: false`
-
-4. Report: "Ralph loop cancelled."
-
-The Stop hook will detect the cancel signal on the next iteration and exit cleanly.
+1. If `.ralph/state/ralph-state.json` missing → "No active Ralph loop found."
+2. If `active: false` → "Ralph loop is already inactive."
+3. Write `.ralph/state/cancel-signal-state.json` (`{}`) — the Stop hook detects this and exits
+4. Set `active: false` in `.ralph/state/ralph-state.json`
+5. Report: "Ralph loop cancelled."
diff --git a/plugins/ralph/skills/ralph/SKILL.md b/plugins/ralph/skills/ralph/SKILL.md
index 90b6cbef..c8edd817 100644
--- a/plugins/ralph/skills/ralph/SKILL.md
+++ b/plugins/ralph/skills/ralph/SKILL.md
@@ -1,90 +1,29 @@
 ---
 name: ralph
-description: PRD-driven persistence loop — keeps Claude working until all user stories pass verification
+description: PRD-driven persistence loop — keeps Claude working until all user stories pass
 ---
 
 # Ralph Loop
 
-You are executing the Ralph persistence loop. Your job is to implement the task completely.
-The Stop hook will keep you running until you write the cancel signal file.
+Stop hook keeps you running until you write the cancel signal.
 
-## Activation (first run only)
+## Activation
+1. `mkdir -p .ralph/state/`
+2. Write `.ralph/state/ralph-activating` with the task description
 
-1. Parse the task from the invocation: `/ralph "your task description"`
-2. Create `.ralph/state/` directory if it does not exist
-3. Write `.ralph/state/ralph-activating` with the task description as content
-4. Proceed to PRD writing
+## PRD (`--no-prd` to skip → auto-generate single-story prd)
+Create `.ralph/prd.json`: `project`, `userStories[]` each with `id`, `title`, `acceptanceCriteria[]`, `priority`, `passes:false`. Small testable stories, dependency-ordered.
 
-## PRD Writing (skip with `--no-prd`)
+## Loop
+1. Read `.ralph/progress.txt` if it exists
+2. Find highest-priority `passes:false` story
+3. All stories pass → go to Done
+4. Implement one story at a time, run tests
+5. Pass → set `passes:true`. Fail → append `[ITER N] US-XXX: <reason>` to `.ralph/progress.txt`
 
-Create `.ralph/prd.json` with this structure:
-
-```json
-{
-  "project": "<project name>",
-  "description": "<task description>",
-  "userStories": [
-    {
-      "id": "US-001",
-      "title": "<story title>",
-      "description": "<what the user wants>",
-      "acceptanceCriteria": ["<testable criterion>", "..."],
-      "priority": 1,
-      "passes": false
-    }
-  ]
-}
-```
-
-Rules:
-
-- Each story must have clear, testable acceptance criteria
-- Order stories by dependency (foundational first)
-- Keep stories small — one story = one focused piece of functionality
-
-## Story Execution Loop
-
-On each iteration:
-
-1. Read `.ralph/progress.txt` for learnings from previous iterations
-2. Read `.ralph/prd.json` and find the highest-priority story with `passes: false`
-3. If all stories have `passes: true`, proceed to Completion Verification
-4. Implement the story following TDD: write failing test → implement → make pass
-5. Run the project's test suite
-6. If tests pass: set `passes: true` for this story in `prd.json`
-7. If tests fail: append learnings to `.ralph/progress.txt`:
-
-   ```text
-   [ITERATION N] Story US-XXX failed: <what went wrong> / <what to try next>
-   ```
-
-## Completion Verification
-
-When all stories have `passes: true`:
-
-1. **Architect review** (skip with `--critic=none`): Review the full implementation for design quality,
-   edge cases, and code clarity. Fix any issues found.
-2. **Deslop pass** (skip with `--no-deslop`): Remove AI-generated boilerplate, overly verbose comments,
-   unnecessary abstractions, and any code that exists for no clear reason.
-3. **Regression test run**: Run the full test suite. All tests must pass.
-
-## Completion
-
-When verification passes:
-
-1. Write `.ralph/state/cancel-signal-state.json` (content can be empty `{}`)
-2. Output: `<promise>COMPLETE</promise>`
-
-The Stop hook will detect the cancel signal and exit the loop.
-
-## Flags
-
-- `--no-prd` — Skip PRD writing, treat the task description as the single story
-- `--no-deslop` — Skip the deslop pass
-- `--critic=none` — Skip architect review (default: architect)
-
-## Important
-
-- Never declare completion without writing the cancel signal file
-- Never skip the regression test run
-- Read `progress.txt` at the start of every iteration — past failures contain critical information
+## Done
+1. Review code quality (`--critic=none` to skip)
+2. Remove slop: unnecessary comments, dead code, over-abstractions (`--no-deslop` to skip)
+3. Run full test suite — must pass
+4. Write `.ralph/state/cancel-signal-state.json` (`{}`)
+5. Reply `<promise>COMPLETE</promise>`