rails_console_ai 0.31.0 → 0.33.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: '083adc4ffd8b69f950fbc4ab18f2ac056e8642c6946c8ba46e67e19e4299e0f2'
-  data.tar.gz: a01fef0fe45dfa157d638a7d4edb1ee5c8d9753be3a79956b3ebde220d3ee939
+  metadata.gz: 8af25f720aae1fe703b947bdc860dcace9a13e2e1db61d9d698634fa46838e59
+  data.tar.gz: efc3b4c00ea66d941e5fb03c61543398fed5ffdde700c44c080ee5e8c8e09ad6
 SHA512:
-  metadata.gz: 56b8fdcd0cf445d9480d4e56e252271241af017c8ca51e2a11af36d2158ce4a3a1fdf9670685f931b8cbd0994e6828e6bac08d10247e8bec1af8c66ad83c47f4
-  data.tar.gz: 6ccb72b8b8e9b157b28722d1e532d2abf5df8037f4ba4f1f3635bb37ad8a3441e48d832d903f6f92b6dce80106082215368dde15626eb0fc41d93da60100f261
+  metadata.gz: 0b8de788b6391a303815cd160fb57ce4300c33dfd3f66645ce31f9bb256de0922b7ce8d26fd05c23dc8802c80da6201ffb9935c89cfd82b94ebde609b1631733
+  data.tar.gz: 5c73d678909c5e950cff5191b506a72570c2eb50bd2d6f01602bfe63ec2fade56c94b88d91f6d487803b073abd7b3565c9b92c355f90f7e32be82e465b72b7f9

data/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 All notable changes to this project will be documented in this file.
 
+## [0.33.0]
+
+- Support Claude Opus 5 and make it the default model for the Anthropic and Bedrock providers
+- Add runaway-loop circuit breakers to the tool loop — break when the same error signature repeats (warn at 3 occurrences, stop at 5) or when a token budget is exceeded (nudge at 500K input tokens, force a final answer at 1M), with a tailored wrap-up prompt for each break reason
+- Fail fast on known environment errors: a built-in hint recognizes decryption key misconfiguration (`bad decrypt` / `ActiveRecord::Encryption`) and tells the model not to retry, and apps can register their own hints via `config.error_hints`
+
+## [0.32.0]
+
+- Switch default models to Claude Sonnet 5 and Opus 4.8, and rework pricing to match models by family so dated snapshots and Bedrock IDs are covered
+- Add an abort mechanism for background sub-agents
+- Require human approval before AI-proposed memory changes take effect
+- Keep `explore_output` focused on exploring a previous command's output instead of spinning off into database reads
+- Fix loading `ApplicationController` under the classic autoloader
+
 ## [0.31.0]
 
 - Simplify DB-backed skills, sub-agents, and memories to a single content field per record, replacing the separate frontmatter/body columns and streamlining the models, controllers, and forms
@@ -22,32 +36,6 @@ All notable changes to this project will be documented in this file.
 - Harden `Skill` / `Agent` model accessors to return safe defaults when newly added columns are not yet present
 - Fix built-in agent `.md` files with UTF-8 characters failing to load under US-ASCII locales
 
-## [Unreleased]
-
-- New skill / memory / agent pages now have a "Paste a .md file" textarea at the top. Paste the contents of a Markdown file with YAML frontmatter (same format used by `.rails_console_ai/{skills,memories,agents}/*.md` and the gem's built-in agents), click "Parse pasted content ↓", and the form fields below are prefilled from the parse. You still click Create to actually save, so the normal proposed-status + version-row flow applies — and you can tweak the parsed content before saving. Useful for moving an existing on-disk record into the versioned DB store
-- `RailsConsoleAi::SkillLoader.parse`, `RailsConsoleAi::AgentLoader.parse`, and `RailsConsoleAi::Tools::MemoryTools.parse` are now public class methods. They return a parsed frontmatter+body hash, or `nil` on malformed input
-
-- Usage tracking for DB-backed skills, memories, and agents. New `use_count` and `last_used_at` columns are bumped atomically (one SQL `UPDATE … SET use_count = use_count + 1, last_used_at = NOW()`, no callbacks, no `updated_at` change) when:
-  - `activate_skill` resolves a DB skill
-  - `recall_memory` resolves a DB memory, or `recall_memories` returns a DB memory in its result set
-  - `delegate_task` resolves a DB sub-agent
-  File / built-in records have no DB row so they're not tracked (the web UI shows `—` for them). System-prompt summary inclusion does **not** count — counters only move when the AI actively invokes/loads the record. Surfaced in the index tables (with a "Sort: most used" toggle) and on each show page (Times activated / Times recalled / Times invoked, plus Last used)
-- Bugfix: `RailsConsoleAi.migrate!` now always re-runs the `setup_*_tables!` methods so column-add probes execute on upgrades. Previously, when the base table already existed, the outer `unless table_exists?` guard skipped column probes entirely, leaving new columns un-added and causing `NameError: undefined local variable or method 'status'` from `Skill#proposed?` on host apps that hadn't fully migrated
-- Bugfix: `Skill` / `Agent` model accessors for `status` / `approved_by` / `approved_at` / `use_count` / `last_used_at` are now defensive — they return safe defaults via `has_attribute?` rather than raising NameError when the column hasn't been added yet. The `status` inclusion validation is also gated on the column being present
-
-- Sub-agents can now be stored in the database, versioned, and approved through the web UI — same workflow as skills. DB-backed agents start as `proposed` and are invisible to `delegate_task` until a human clicks Approve at `/rails_console_ai/agents`; editing an approved agent reverts it to proposed. Built-in (gem-shipped) and file-based agents are pre-approved and continue working unchanged
-- New AI tools `save_agent` and `delete_agent` — the AI can now draft a sub-agent definition (lands as proposed, awaits human approval) and request deletion. `delegate_task` now emits a specific "awaiting human approval" error when the AI references a proposed DB agent
-- New web UI section at `/rails_console_ai/agents` with three-source listing (DB / FILE / BUILTIN badges) plus the full skills-style CRUD + version history + diff + restore + approve. Built-in agents are read-only with a "Create DB override" link that prefills the new-agent form
-- `ai_db_setup` / `ai_db_migrate` now also create `rails_console_ai_agents` and `rails_console_ai_agent_versions` tables idempotently
-- Fix: built-in agent .md files containing UTF-8 (em-dashes, smart quotes) now load correctly. `safe_load_builtin_agents` was using `File.read`'s locale-default encoding, which silently swallowed `Encoding::CompatibilityError` on US-ASCII locales
-
-- Skills and memories can now be stored in the database (in addition to the existing on-disk `.rails_console_ai/skills` and `.rails_console_ai/memories` files). DB-backed records are versioned — every save creates a `SkillVersion` / `MemoryVersion` row with `edited_by` and an optional change note
-- DB-backed skills start in a **proposed** state and cannot be activated by the AI until a human approves them in the web UI. Editing an approved skill reverts it to proposed. File-backed skills are unaffected (already git-tracked, considered pre-approved). Memories are not gated
-- `SkillLoader#load_all_skills` and `MemoryTools#load_all_memories` now return the union of DB and file records (DB wins on name collision). `SkillLoader#find_skill` / `skill_summaries` filter out proposed skills so the AI never sees them
-- `save_skill` / `save_memory` AI tools accept an optional `target` parameter (`"db"` default, `"file"` to write to disk) plus `change_note`. When the AI saves a DB skill, the tool response tells it the skill is awaiting human approval
-- New web UI sections at `/rails_console_ai/skills` and `/rails_console_ai/memories` provide list, view, create, edit, delete, version history, side-by-side diff, and restore. Skills also have an Approve button and PROPOSED / APPROVED badges. File-sourced records are surfaced but read-only in the UI
-- `ai_db_setup` / `ai_db_migrate` now create the new `rails_console_ai_skills`, `rails_console_ai_skill_versions`, `rails_console_ai_memories`, and `rails_console_ai_memory_versions` tables idempotently, plus the `status` / `approved_by` / `approved_at` columns on skills
-
 ## [0.29.0]
 
 - Allow steering Slack conversations mid-run by sending follow-up messages that are folded in as user guidance at the next tool-loop boundary

data/README.md

@@ -291,7 +291,7 @@ RailsConsoleAi.configure do |config|
 end
 ```
 
-Default model: `claude-sonnet-4-6`. Thinking model: `claude-opus-4-6`. Prompt caching is enabled automatically.
+Default model: `claude-sonnet-5`. Thinking model: `claude-opus-5`. Prompt caching is enabled automatically.
 
 ### OpenAI
 
@@ -317,8 +317,8 @@ gem 'aws-sdk-bedrockruntime'
 RailsConsoleAi.configure do |config|
   config.provider = :bedrock
   config.bedrock_region = 'us-east-1'
-  # config.model = 'us.anthropic.claude-sonnet-4-6'           # default
-  # config.thinking_model = 'us.anthropic.claude-opus-4-6-v1' # default
+  # config.model = 'us.anthropic.claude-sonnet-5'        # default
+  # config.thinking_model = 'us.anthropic.claude-opus-5'   # default
 end
 ```
 
@@ -365,25 +365,25 @@ Before adopting a new Claude model, smoke-test it against the Anthropic or Bedro
 
 ```bash
 # Anthropic — provider inferred from the `claude-` prefix
-ANTHROPIC_API_KEY=sk-ant-... bin/smoke_model.rb --model claude-opus-4-7
+ANTHROPIC_API_KEY=sk-ant-... bin/smoke_model.rb --model claude-opus-4-8
 
 # Bedrock — provider inferred from the regional `us.anthropic.` prefix.
 # Requires the aws-sdk-bedrockruntime gem and AWS credentials in the environment.
-bin/smoke_model.rb --model us.anthropic.claude-opus-4-7
+bin/smoke_model.rb --model us.anthropic.claude-opus-4-8
 
 # Bedrock in another region
-bin/smoke_model.rb --model eu.anthropic.claude-opus-4-7 --region eu-west-1
+bin/smoke_model.rb --model eu.anthropic.claude-opus-4-8 --region eu-west-1
 
 # Subset of checks, e.g. when iterating on cache behavior
-bin/smoke_model.rb --model claude-sonnet-4-6 --checks cache
+bin/smoke_model.rb --model claude-sonnet-5 --checks cache
 
 # Force a provider when the model ID is ambiguous
-bin/smoke_model.rb --provider anthropic --model claude-opus-4-7
+bin/smoke_model.rb --provider anthropic --model claude-opus-4-8
 ```
 
 `DEBUG=1` enables the providers' raw request/response logging.
 
-If the model rejects a parameter the gem sends by default (e.g. opus-4-7 deprecated `temperature`), add the model ID to `Configuration::MODELS_WITHOUT_TEMPERATURE` in `lib/rails_console_ai/configuration.rb` so the providers omit the field.
+Pricing, default max tokens, and parameter support (e.g. which families reject `temperature`) are keyed by model family in `Configuration::MODEL_FAMILIES` (`lib/rails_console_ai/configuration.rb`). Family matching is by substring, so one entry covers bare IDs, dated snapshots, and Bedrock inference profiles (`us.` / `global.` prefixes). When adopting a new model family, add an entry there and smoke-test it.
 
 ## Configuration
 
@@ -396,9 +396,48 @@ RailsConsoleAi.configure do |config|
   config.timeout = 30                 # HTTP timeout in seconds
   config.max_tool_rounds = 200        # safety cap on tool-use loops
   config.code_search_paths = %w[app]  # directories for list_files / search_code
+
+  # Runaway-loop circuit breakers (see "Runaway sessions" below)
+  config.token_nudge_threshold = 500_000    # input tokens in one tool loop → nudge model to wrap up (nil disables)
+  config.token_stop_threshold  = 1_000_000  # input tokens in one tool loop → force a final answer (nil disables)
 end
 ```
 
+### Runaway sessions & known-issue hints
+
+Two mechanisms stop the LLM from burning tokens on a dead end:
+
+**Known-issue hints** — when executed code fails (or prints a rescued error) matching a
+known environment-level problem, the tool result includes explicit guidance telling the
+model not to retry. The built-in hint recognizes decryption failures
+(`OpenSSL::Cipher::CipherError` / "bad decrypt" / `ActiveRecord::Encryption` errors),
+which indicate a missing or placeholder encryption key in the console process — an
+environment issue no amount of retrying can fix. Repeat occurrences escalate the message.
+Apps can add their own:
+
+```ruby
+RailsConsoleAi.configure do |config|
+  config.error_hints << {
+    name: :vpn_required,
+    pattern: /Errno::ECONNREFUSED.*10\.8\./,
+    hint: "This host is only reachable over the VPN, which this console does not have. " \
+          "Do not retry; report the limitation to the user."
+  }
+end
+```
+
+**Circuit breakers** — inside a single tool loop, the engine tracks:
+
+- *Identical tool calls* (same tool + same args): warns at 3, breaks at 5.
+- *Repeated error signatures* (same normalized error from **different** code): warns at 3,
+  breaks at 5. This catches "new code, same dead end" loops that identical-call detection
+  misses. Digits are normalized so varying record IDs don't disguise the same failure.
+- *Token budget*: at `token_nudge_threshold` cumulative input tokens the model is told to
+  wrap up; at `token_stop_threshold` the loop is stopped and a final answer is forced.
+
+When any breaker trips, the model is asked to summarize what it established, what it
+could not determine and why, and what a human should do next — instead of iterating.
+
 ### Code Search Paths
 
 By default, `list_files` and `search_code` only look in `app/`. If your project has code in other directories (e.g. a frontend in `public/portal`, or shared code in `lib`), add them:
@@ -513,12 +552,17 @@ id = RailsConsoleAi.run_agent("How many users signed up yesterday?", name: 'dail
 # => 4821 (Integer session id, returned immediately)
 
 RailsConsoleAi.check_agent(id)
-# => 'queued' | 'running' | 'ready' | 'failed' | nil
+# => 'queued' | 'running' | 'ready' | 'failed' | 'aborted' | nil
 
 RailsConsoleAi.get_agent_response(id)
 # => { status: 'ready', result: "1,432 users signed up yesterday.\n", error: nil }
+
+RailsConsoleAi.abort_agent(id)
+# => true (aborted) | false (already finished, or unknown id)
 ```
 
+`abort_agent` cancels a run: a queued run is never picked up, and a run already executing keeps going but its result is discarded when it completes -- the session stays `status='aborted'`.
+
 `run_agent` enqueues a row in the sessions table with `mode='agent_api'` and `status='queued'`. A separate long-running rake task picks them up and runs each in its own thread using the same engine that powers `ai "..."` in the console.
 
 ### Per-run options

data/app/controllers/rails_console_ai/agent_versions_controller.rb

@@ -1,5 +1,5 @@
 module RailsConsoleAi
-  class AgentVersionsController < ApplicationController
+  class AgentVersionsController < RailsConsoleAi::ApplicationController
     before_action :load_agent
 
     def index

data/app/controllers/rails_console_ai/agents_controller.rb

@@ -1,7 +1,7 @@
 require 'rails_console_ai/agent_loader'
 
 module RailsConsoleAi
-  class AgentsController < ApplicationController
+  class AgentsController < RailsConsoleAi::ApplicationController
     before_action :load_agent, only: [:show, :edit, :update, :destroy, :approve]
 
     def index

data/app/controllers/rails_console_ai/memories_controller.rb

@@ -1,8 +1,8 @@
 require 'rails_console_ai/tools/memory_tools'
 
 module RailsConsoleAi
-  class MemoriesController < ApplicationController
-    before_action :load_memory, only: [:show, :edit, :update, :destroy]
+  class MemoriesController < RailsConsoleAi::ApplicationController
+    before_action :load_memory, only: [:show, :edit, :update, :destroy, :approve]
 
     def index
       @memories = Tools::MemoryTools.new.load_all_memories
@@ -73,6 +73,26 @@ module RailsConsoleAi
       end
     end
 
+    def approve
+      redirect_to memories_path, alert: file_memory_message and return unless @memory.is_a?(RailsConsoleAi::Memory)
+
+      approver = params[:approved_by].presence ||
+                 (request.respond_to?(:remote_user) && request.remote_user.presence) ||
+                 'web'
+
+      if @memory.approved?
+        redirect_to memory_path(@memory), notice: 'Memory is already approved.'
+        return
+      end
+
+      begin
+        @memory.approve!(approved_by: approver)
+        redirect_to memory_path(@memory), notice: "Approved by #{approver}. The AI can now recall this memory."
+      rescue ArgumentError, ActiveRecord::RecordInvalid => e
+        redirect_to memory_path(@memory), alert: "Could not approve: #{e.message}"
+      end
+    end
+
     def diff
       @memory = Memory.find(params[:memory_id])
       @from = @memory.versions.find(params[:from])

data/app/controllers/rails_console_ai/memory_versions_controller.rb

@@ -1,5 +1,5 @@
 module RailsConsoleAi
-  class MemoryVersionsController < ApplicationController
+  class MemoryVersionsController < RailsConsoleAi::ApplicationController
     before_action :load_memory
 
     def index

data/app/controllers/rails_console_ai/sessions_controller.rb

@@ -1,5 +1,5 @@
 module RailsConsoleAi
-  class SessionsController < ApplicationController
+  class SessionsController < RailsConsoleAi::ApplicationController
     PER_PAGE = 50
 
     def index

data/app/controllers/rails_console_ai/skill_versions_controller.rb

@@ -1,5 +1,5 @@
 module RailsConsoleAi
-  class SkillVersionsController < ApplicationController
+  class SkillVersionsController < RailsConsoleAi::ApplicationController
     before_action :load_skill
 
     def index

data/app/controllers/rails_console_ai/skills_controller.rb

@@ -1,7 +1,7 @@
 require 'rails_console_ai/skill_loader'
 
 module RailsConsoleAi
-  class SkillsController < ApplicationController
+  class SkillsController < RailsConsoleAi::ApplicationController
     before_action :load_skill, only: [:show, :edit, :update, :destroy, :approve]
 
     def index

data/app/models/rails_console_ai/memory.rb

@@ -4,6 +4,10 @@ module RailsConsoleAi
   class Memory < ActiveRecord::Base
     self.table_name = 'rails_console_ai_memories'
 
+    STATUS_PROPOSED = 'proposed'.freeze
+    STATUS_APPROVED = 'approved'.freeze
+    STATUSES = [STATUS_PROPOSED, STATUS_APPROVED].freeze
+
     has_many :versions,
              -> { order(created_at: :desc) },
              class_name: 'RailsConsoleAi::MemoryVersion',
@@ -12,11 +16,14 @@ module RailsConsoleAi
 
     validates :content, presence: true
     validates :name, presence: true, uniqueness: { case_sensitive: false }
+    validates :status, inclusion: { in: STATUSES }
     validate  :content_parses
 
     before_validation :sync_name_from_content
 
     scope :alphabetical, -> { order(Arel.sql('LOWER(name)')) }
+    scope :approved, -> { where(status: STATUS_APPROVED) }
+    scope :proposed, -> { where(status: STATUS_PROPOSED) }
 
     def self.connection
       klass = RailsConsoleAi.configuration.connection_class
@@ -42,6 +49,9 @@ module RailsConsoleAi
     def description; parsed['description']; end
     def tags;        Array(parsed['tags']); end
 
+    def proposed?; status.to_s == STATUS_PROPOSED; end
+    def approved?; status.to_s == STATUS_APPROVED; end
+
     def self.record_use!(id)
       where(id: id).update_all([
         'use_count = COALESCE(use_count, 0) + 1, last_used_at = ?',
@@ -60,6 +70,9 @@ module RailsConsoleAi
         'description'  => description,
         'tags'         => tags,
         'content'      => content,
+        'status'       => status,
+        'approved_by'  => approved_by,
+        'approved_at'  => approved_at,
         'use_count'    => use_count,
         'last_used_at' => last_used_at,
         'source'       => :db,
@@ -67,14 +80,25 @@ module RailsConsoleAi
       }
     end
 
-    def update_with_version!(attrs, edited_by: nil, change_note: nil)
+    # Assigns attrs, saves, and records one MemoryVersion snapshot.
+    # Any change to `content` reverts approval back to "proposed" unless
+    # `preserve_approval: true` is passed (approve! does this).
+    def update_with_version!(attrs, edited_by: nil, change_note: nil, preserve_approval: false)
       transaction do
         assign_attributes(attrs)
+
+        if !preserve_approval && approved? && changes.key?('content')
+          self.status      = STATUS_PROPOSED
+          self.approved_by = nil
+          self.approved_at = nil
+        end
+
         save!
         RailsConsoleAi::MemoryVersion.create!(
           memory_id:   id,
           name:        name,
           content:     content,
+          status:      status,
           edited_by:   edited_by,
           change_note: change_note
         )
@@ -82,6 +106,21 @@ module RailsConsoleAi
       self
     end
 
+    def approve!(approved_by:)
+      raise ArgumentError, 'approved_by is required' if approved_by.to_s.strip.empty?
+
+      update_with_version!(
+        {
+          status:      STATUS_APPROVED,
+          approved_by: approved_by,
+          approved_at: Time.now.utc
+        },
+        edited_by: approved_by,
+        change_note: "Approved by #{approved_by}",
+        preserve_approval: true
+      )
+    end
+
     private
 
     def sync_name_from_content

data/app/views/rails_console_ai/memories/index.html.erb

@@ -11,6 +11,13 @@
   </form>
 </div>
 
+<% proposed_count = @memories.count { |m| m['source'] == :db && m['status'] != 'approved' } %>
+<% if proposed_count > 0 %>
+  <div class="flash flash-alert" style="margin-bottom:16px;">
+    <strong><%= proposed_count %></strong> memor<%= proposed_count == 1 ? 'y' : 'ies' %> awaiting approval. The AI cannot recall them until a human approves.
+  </div>
+<% end %>
+
 <% if @memories.empty? %>
   <div class="meta-card">
     <p class="text-muted">No memories yet. <%= link_to 'Create one', new_memory_path %> or drop a Markdown file in <code>.rails_console_ai/memories/</code>.</p>
@@ -22,7 +29,7 @@
         <th>Name</th>
         <th>Description</th>
         <th>Tags</th>
-        <th>Source</th>
+        <th>Source / Status</th>
         <th title="Number of times the AI recalled this memory (DB only)">Uses</th>
         <th>Last used</th>
         <th></th>
@@ -40,6 +47,11 @@
           <td>
             <% if m['source'] == :db %>
               <span class="badge source-db">DB</span>
+              <% if m['status'] == 'approved' %>
+                <span class="badge status-approved">APPROVED</span>
+              <% else %>
+                <span class="badge status-proposed">PROPOSED</span>
+              <% end %>
             <% else %>
               <span class="badge source-file">FILE</span>
             <% end %>

data/app/views/rails_console_ai/memories/show.html.erb

@@ -8,11 +8,23 @@
 
 <%= link_to '← All memories', memories_path, class: 'back-link' %>
 
+<% if is_db && !@memory.approved? %>
+  <div class="flash flash-alert" style="margin-bottom:16px;">
+    <strong>Awaiting approval.</strong> The AI cannot recall this memory until a human approves it.
+    Review the content below, then click <strong>Approve</strong>.
+  </div>
+<% end %>
+
 <div class="meta-card">
   <h2 style="margin-bottom:8px;">
     <%= name %>
     <% if is_db %>
       <span class="badge source-db">DB</span>
+      <% if @memory.approved? %>
+        <span class="badge status-approved">APPROVED</span>
+      <% else %>
+        <span class="badge status-proposed">PROPOSED</span>
+      <% end %>
     <% else %>
       <span class="badge source-file">FILE</span>
     <% end %>
@@ -24,6 +36,10 @@
       <div class="meta-item"><label>Versions</label><span><%= @memory.versions.count %></span></div>
       <div class="meta-item"><label>Times recalled</label><span><%= @memory.use_count %></span></div>
       <div class="meta-item"><label>Last recalled</label><span><%= @memory.last_used_at&.strftime('%Y-%m-%d %H:%M') || 'never' %></span></div>
+      <% if @memory.approved? %>
+        <div class="meta-item"><label>Approved by</label><span><%= @memory.approved_by || '—' %></span></div>
+        <div class="meta-item"><label>Approved at</label><span><%= @memory.approved_at&.strftime('%Y-%m-%d %H:%M') || '—' %></span></div>
+      <% end %>
     <% else %>
       <div class="meta-item"><label>Path</label><span class="mono"><%= file_key %></span></div>
     <% end %>
@@ -32,6 +48,12 @@
 
 <div class="btn-bar">
   <% if is_db %>
+    <% unless @memory.approved? %>
+      <%= form_with url: approve_memory_path(@memory), method: :post, local: true, style: 'display:inline' do %>
+        <input type="text" name="approved_by" placeholder="Your name (required)" style="padding:7px 10px; border:1px solid #ccc; border-radius:4px; margin-right:6px;">
+        <button type="submit" class="btn btn-danger" data-confirm="Approve this memory? The AI will then be able to recall it.">Approve</button>
+      <% end %>
+    <% end %>
     <%= link_to 'Edit', edit_memory_path(@memory), class: 'btn' %>
     <%= link_to 'Versions', memory_versions_path(@memory), class: 'btn btn-secondary' %>
     <%= button_to 'Delete', memory_path(@memory), method: :delete, class: 'btn btn-danger', form: { style: 'display:inline' },data: { confirm: 'Delete this memory? Past versions remain in history.' } %>
@@ -49,6 +71,7 @@
         <div>
           <strong>v<%= v.id %></strong>
           <span class="ver-meta">— <%= v.created_at.strftime('%Y-%m-%d %H:%M') %> by <%= v.edited_by || 'unknown' %></span>
+          <% if v.status == 'approved' %><span class="badge status-approved">APPROVED</span><% end %>
           <% if v.change_note.present? %><div class="ver-meta"><%= v.change_note %></div><% end %>
         </div>
         <div>

data/config/routes.rb

@@ -17,6 +17,9 @@ RailsConsoleAi::Engine.routes.draw do
   end
 
   resources :memories do
+    member do
+      post :approve
+    end
     collection do
       get :diff
     end

data/lib/generators/rails_console_ai/templates/initializer.rb

@@ -5,11 +5,11 @@ RailsConsoleAi.configure do |config|
   # API key (or set ANTHROPIC_API_KEY / OPENAI_API_KEY env var)
   # config.api_key = 'sk-...'
 
-  # Model override (defaults: claude-opus-4-6 for Anthropic, gpt-5.3-codex for OpenAI)
-  # config.model = 'claude-opus-4-6'
+  # Model override (defaults: claude-sonnet-5 for Anthropic, gpt-5.3-codex for OpenAI)
+  # config.model = 'claude-opus-4-8'
 
-  # Max tokens for LLM response
-  config.max_tokens = 4096
+  # Max tokens for LLM response (default resolves per model family; 16000 for Claude models)
+  # config.max_tokens = 4096
 
   # Temperature (0.0 - 1.0)
   config.temperature = 0.2
@@ -35,7 +35,7 @@ RailsConsoleAi.configure do |config|
   # AWS Bedrock provider (uses AWS credential chain — no API key needed):
   # config.provider = :bedrock
   # config.bedrock_region = 'us-east-1'
-  # config.model = 'us.anthropic.claude-sonnet-4-6'
+  # config.model = 'us.anthropic.claude-sonnet-5'
 
   # Debug mode: prints full API requests/responses and tool calls to stderr
   # config.debug = true

data/lib/rails_console_ai/agent_runner.rb

@@ -134,6 +134,10 @@ module RailsConsoleAi
       end
       result_text = compose_result(channel.captured_output, exec_result)
 
+      if aborted?(session.id)
+        puts ">> aborted (result discarded)"
+        return
+      end
       SessionLogger.update(session.id, status: 'ready', result: result_text)
 
       elapsed = ((Process.clock_gettime(Process::CLOCK_MONOTONIC) - started) * 1000).round
@@ -142,16 +146,23 @@ module RailsConsoleAi
       puts ">> ready (#{elapsed}ms) #{preview}"
     rescue RunnerTimeoutError => e
       warn ">> TIMEOUT #{e.message}"
-      SessionLogger.update(session.id, status: 'failed', error_message: e.message)
+      SessionLogger.update(session.id, status: 'failed', error_message: e.message) unless aborted?(session.id)
     rescue => e
       warn ">> FAILED #{e.class}: #{e.message}"
       e.backtrace&.first(5)&.each { |line| warn "   #{line}" }
+      return if aborted?(session.id)
       SessionLogger.update(session.id,
         status: 'failed',
         error_message: "#{e.class}: #{e.message}\n#{Array(e.backtrace).first(10).join("\n")}"
       )
     end
 
+    # An abort (RailsConsoleAi.abort_agent) flips the row to 'aborted' while we
+    # run; in that case the run's outcome must not overwrite the abort.
+    def aborted?(session_id)
+      Session.where(id: session_id).pluck(:status).first == 'aborted'
+    end
+
     # When cap is nil or non-positive, run inline. Otherwise spawn a nested
     # worker thread, join with the deadline, and kill + raise on overshoot.
     # Kept localized (vs Timeout.timeout) so a runaway provider call can't

data/lib/rails_console_ai/channel/console.rb

@@ -368,7 +368,7 @@ module RailsConsoleAi
         config = RailsConsoleAi.configuration
         model = @engine.effective_model
         thinking = config.resolved_thinking_model
-        pricing = Configuration::PRICING[model]
+        pricing = Configuration.pricing_for(model)
 
         @real_stdout.puts "\e[36m  Model info:\e[0m"
         @real_stdout.puts "\e[2m    Provider:        #{config.provider}\e[0m"