feat!: replace Experience Engine with repo-local troubleshooting memory (closes #200) (#203)

* refactor: relocate redact.ts to util/ (carve-out before Experience Engine removal)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* refactor: extract lean observability instrumentation seam from experience/telemetry

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat!: remove Experience Engine modules (compaction/telemetry/ghost/sharing) — GH #200

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* build: rebuild cdp-bridge dist after Experience Engine removal

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat!: remove rn-agent-{compact,health,export,import} commands + experience-engine scaffold

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat: scaffold .rn-agent/local/ troubleshooting doc at SessionStart

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* feat: capture redacted tool failures to .rn-agent/local/session-buffer.jsonl

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix: restore project-gated capture guard; test sets up project evidence

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* fix(security): redaction parity with redact.ts — ${1} keyed rule, redact-before-truncate, slurp-mode multiline, generic-prefix+PEM patterns

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat: gated Stop hook synthesizes session buffer into troubleshooting.md

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* feat: inject repo-local troubleshooting doc at SessionStart

When .rn-agent/local/troubleshooting.md exists with real content (beyond
the scaffold template), the detect-rn-project.sh SessionStart hook now
injects that content into the agent context before the main banner, so
the agent begins each session aware of repo-specific config and known
gotchas. Also corrects banner tool count from 75 → 76.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* docs: point agents at .rn-agent/local/troubleshooting.md; changeset; close GH #200

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* chore: drop stale experience-engine reference in nav-graph comment

* test: drop engine-only unit tests; repoint observer test to observability/instrumentation

The Experience Engine removal orphaned 4 unit tests importing dist/experience/*.
Delete the 3 that tested deleted code (audit-b4-experience, gh-106-flow-bundle,
gh-106-sharing-flows); repoint observability-observer to the relocated
instrumentation seam. Full unit suite green (1590/1590).

* docs: remove published docs for deleted rn-agent-* commands; point at troubleshooting memory

* docs: add Troubleshooting Memory feature page + sidebar entry (GH #200)

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: Lykhoyda

.changeset/repo-local-troubleshooting-memory.md

@@ -0,0 +1,12 @@
+---
+"rn-dev-agent-plugin": minor
+---
+
+Replace the Experience Engine with a repo-local troubleshooting memory.
+
+`/rn-agent-compact`, `/rn-agent-health`, `/rn-agent-export`, and `/rn-agent-import`
+are removed (GH #200: compaction had no runnable entry point and the read path was
+vestigial). In their place, rn-dev-agent now maintains a gitignored
+`.rn-agent/local/troubleshooting.md` per repo: failures are captured by a hook,
+the agent synthesizes them into the doc at session end, and the doc is injected at
+session start so the agent learns this repo's config and gotchas.

.claude-plugin/plugin.json

@@ -42,10 +42,6 @@
     "./commands/build-and-test.md",
     "./commands/rn-feature-dev.md",
     "./commands/proof-capture.md",
-    "./commands/rn-agent-compact.md",
-    "./commands/rn-agent-export.md",
-    "./commands/rn-agent-import.md",
-    "./commands/rn-agent-health.md",
     "./commands/nav-graph.md",
     "./commands/send-feedback.md",
     "./commands/setup.md",

CLAUDE-MD-TEMPLATE.md

@@ -11,6 +11,21 @@ This project uses the **rn-dev-agent** plugin for React Native development and t
 It provides MCP tools across three categories: CDP introspection, device control, and testing.
 Run `/rn-dev-agent:check-env` to verify the current plugin version and tool count.
 
+### 🧠 Repo-local troubleshooting memory
+
+This project keeps an auto-maintained, gitignored notes file at
+`.rn-agent/local/troubleshooting.md` with two sections: **Configuration & How-To**
+(repo-specific facts — Metro start dir, store exposure, testID conventions,
+auth/deeplink, build quirks) and **Troubleshooting** (failure→resolution gotchas).
+
+- **Read it first.** At the start of any device/CDP task, consult this file (the
+  SessionStart hook also injects it) so you don't re-derive known gotchas or
+  re-hit a known failure.
+- **It updates itself.** When rn-dev-agent tool calls fail, a hook records them;
+  at session end a Stop hook asks you to merge new gotchas into this file. If
+  prompted, do it — keep entries concise and under ~2000 tokens total.
+- It is per-developer and never committed.
+
 ### 🚨 MANDATORY PRE-FLIGHT (before ANY device_* call)
 
 Run this 3-step checklist at the start of every UI-touching task. This is the
@@ -96,7 +111,6 @@ afterward — discovery is a one-time cost, replay is the steady state.
 | **Run** | `/rn-dev-agent:run-action <id> [-e KEY=VALUE …]` (calls `cdp_run_action`) | Replays with safety pre-flights (mutates flag, appId match, parameter coverage) and auto-repair on `SELECTOR_NOT_FOUND` |
 | **Self-heal** | `cdp_repair_action <id>` | Fuzzy-matches the stale testID against the live snapshot, patches the YAML in place, bumps `revision`, demotes `status` to `experimental` until next clean replay. Bounded: max 3 attempts/24h, refuses on human edits (mtime check) |
 | **Assert state** | `expect_redux`, `expect_route`, `expect_visible_by_testid`, `expect_text` | Macro-Asserts — embed internal-state assertions inside replays. Maestro asserts pixels; these assert what the app actually believes |
-| **Compact** | `/rn-dev-agent:rn-agent-compact` | Periodic corpus health report — flags cold (90+ day), flaky (>50% failure), or high-churn (5+ repairs/30d) actions. Deletion is human-in-the-loop |
 
 **Canonical loop.** Record a verified walk once → save as an action → in the
 next session, `list-learned-actions` surfaces it for the agent → `run-action`

CLAUDE.md

@@ -64,13 +64,8 @@ Actions (replayable flows — see "Actions" section below):
 /rn-dev-agent:run-action <name> -e K=V — Replay a saved action; auto-repair-aware
 ```
 
-Experience Engine (cross-session learning):
-```
-/rn-dev-agent:rn-agent-health          — Health check the local experience store
-/rn-dev-agent:rn-agent-compact         — Compact telemetry on disk
-/rn-dev-agent:rn-agent-export          — Export the experience store
-/rn-dev-agent:rn-agent-import          — Import an experience snapshot
-```
+Repo-local troubleshooting memory (replaces the Experience Engine):
+`.rn-agent/local/troubleshooting.md` — gitignored, agent-read at SessionStart, auto-updated by the Stop hook after failures (GH #200).
 
 ### How it works
 1. Always start with `cdp_status` — this connects to your running app via CDP

README.md

@@ -100,7 +100,7 @@ An **action** is a saved Maestro flow the agent **emits** when `/test-feature` v
 | `/rn-dev-agent:nav-graph` | Extract and inspect the app navigation graph |
 | `/rn-dev-agent:send-feedback` | Open a GitHub issue with sanitized environment context |
 
-**Experience Engine** (cross-session learning): `/rn-dev-agent:rn-agent-export`, `/rn-dev-agent:rn-agent-import`, `/rn-dev-agent:rn-agent-health`, `/rn-dev-agent:rn-agent-compact` — see [docs](https://lykhoyda.github.io/rn-dev-agent/commands/).
+Repo-local troubleshooting memory: the agent maintains a gitignored `.rn-agent/local/troubleshooting.md` per project (auto-captured failures + config notes), read at session start and updated at session end.
 
 ## What makes this different
 

commands/rn-agent-compact.md

@@ -45,6 +45,7 @@ export default defineConfig({
         { label: 'Getting Started', slug: 'getting-started' },
         { label: 'Architecture', slug: 'architecture' },
         { label: 'Actions', slug: 'actions' },
+        { label: 'Troubleshooting Memory', slug: 'troubleshooting-memory' },
         {
           label: 'Commands',
           items: [
@@ -75,15 +76,6 @@ export default defineConfig({
                 { label: 'nav-graph', slug: 'commands/nav-graph' },
               ],
             },
-            {
-              label: 'Experience Engine',
-              items: [
-                { label: 'rn-agent-health', slug: 'commands/rn-agent-health' },
-                { label: 'rn-agent-compact', slug: 'commands/rn-agent-compact' },
-                { label: 'rn-agent-export', slug: 'commands/rn-agent-export' },
-                { label: 'rn-agent-import', slug: 'commands/rn-agent-import' },
-              ],
-            },
             { label: 'send-feedback', slug: 'commands/send-feedback' },
           ],
         },