brainiac-discord 0.0.11 → 0.0.14

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: e8d2bbbf0efaa83da8904137abf18a256f79e5073fe5896390c0afa1f880856f
-  data.tar.gz: 9b9a19c746dd57d13b232a6864918b99acf9a54128213dac2ff1c0328b90e242
+  metadata.gz: f90a4cf177bc62d55f5773e21f3b5e05aaefbdbff94bae975fcbcb21ae3dbb64
+  data.tar.gz: a9dc84c5ab058a8173c96a6bda909a5ea8959a36fe61065d316b0b408ab2eb54
 SHA512:
-  metadata.gz: 8af3d015077e894932bf05ae2940b8bc9d69895874b88ed548e51c01469a2257bc2bda5dd162a601c775c4745567c257a16c0091b52f5ae4854f4ebb42fecfd3
-  data.tar.gz: bbda431894c7453f4e6a8f9b8c6747884cc9cf39c01a0b812dd7f9a812a5afff63fbbc0f2171ccee88c7deabfd3eeac7216910efbb01b24def1d4d3270ad4db0
+  metadata.gz: c7df8598d6a702644e25c18c76c49aa8f406fa3c0806b2ba9a4eccf0859615fffc027e056d66d2ac42b4d737fdb7b7c00f62e2c2b629a4ee4e4e1adfa9844030
+  data.tar.gz: f7e4ddc91b9d0b819f82000e276f9490fa4fdc8454fd475ef49d09de5d58b917e988be95df2ec8577a4d55cca4ec186cc4cc1d486022912840e49767a41371ee

data/lib/brainiac/plugins/discord/message.rb

@@ -558,8 +558,24 @@ module Brainiac
               )
             end
 
+            # Resolve CLI overrides and work_dir BEFORE building the prompt so we can determine
+            # whether resume will actually work. This prevents sending a lean prompt when
+            # the CLI won't actually have session state.
+            work_dir = chat_mode_fallback(agent_key, agent_name, message_id, chat_mode, thread_worktree_path) ||
+                       (project_config ? project_config["repo_path"] : Dir.pwd)
+            model, effort, cli_provider_override = resolve_overrides(
+              clean_content, project_config, thread_model, thread_effort, thread_cli_provider
+            )
+            resolved = resolve_project_cli_config(project_config || DEFAULT_PROJECT,
+                                                  cli_provider_override: cli_provider_override, agent_name: agent_name)
+
+            # Convert should_resume to the actual flag/symbol. If resolve_resume returns false,
+            # we know the CLI won't actually resume, so we shouldn't send a lean prompt.
+            actual_resume = should_resume ? resolve_resume(true, resolved, work_dir) : false
+            use_lean_prompt = actual_resume && thread_worktree_path
+
             prompt = build_prompt(
-              should_resume: should_resume, thread_worktree_path: thread_worktree_path,
+              should_resume: use_lean_prompt, thread_worktree_path: thread_worktree_path,
               clean_content_for_prompt: clean_content_for_prompt,
               discord_user: discord_user, channel_name: channel_info&.dig("name") || channel_id, reply_context: reply_context,
               channel_history: channel_history, thread_root_context: thread_root_context,
@@ -567,15 +583,9 @@ module Brainiac
               brain_context: brain_context, agent_name: agent_name, supersede_key: supersede_key
             )
 
-            work_dir = chat_mode_fallback(agent_key, agent_name, message_id, chat_mode, thread_worktree_path) ||
-                       (project_config ? project_config["repo_path"] : Dir.pwd)
             prompt_file = File.join(response_dir, "discord-prompt-#{timestamp}-#{agent_key}-#{message_id}.md")
             File.write(prompt_file, prompt)
 
-            model, effort, cli_provider_override = resolve_overrides(
-              clean_content, project_config, thread_model, thread_effort, thread_cli_provider
-            )
-
             persist_overrides(thread_map_key, clean_content, project_config,
                               cli_provider: cli_provider_override, model: model, effort: effort,
                               prev_cli_provider: thread_cli_provider, prev_model: thread_model, prev_effort: thread_effort)
@@ -592,8 +602,8 @@ module Brainiac
               agent_key: agent_key, agent_name: agent_name, bot_token: bot_token,
               channel_id: channel_id, message_id: message_id, discord_user: discord_user,
               work_dir: work_dir, prompt_file: prompt_file, response_file: response_file,
-              meta_file: meta_file, model: model, effort: effort, should_resume: should_resume,
-              cli_provider_override: cli_provider_override, project_config: project_config,
+              meta_file: meta_file, model: model, effort: effort, actual_resume: actual_resume,
+              resolved: resolved, project_config: project_config,
               session_key: session_key, supersede_key: supersede_key,
               attachment_paths: attachment_paths, timestamp: timestamp, response_dir: response_dir
             )
@@ -816,6 +826,13 @@ module Brainiac
             debounced_repo_fetch(repo_path)
             worktree_path = create_or_reuse_worktree(repo_path: repo_path, branch: branch)
 
+            # Register a work item so brainiac-github can route PR comments/reviews back
+            # to the agent that actually owns this Discord thread's worktree. Without this,
+            # PRs created from a conversational session have no work item and comments
+            # fall back to the project's default agent.
+            project_key = PROJECTS.find { |_k, v| v == project_config }&.first
+            register_work_item(branch: branch, worktree: worktree_path, project: project_key, agent: agent_name)
+
             cli, model, effort = detect_thread_overrides(
               project_config, clean_content,
               fallback_cli: existing&.dig("cli_provider"),
@@ -886,17 +903,15 @@ module Brainiac
 
           def spawn_agent(agent_key:, agent_name:, bot_token:, channel_id:, message_id:, discord_user:,
                           work_dir:, prompt_file:, response_file:, meta_file:, model:, effort:,
-                          should_resume:, cli_provider_override:, project_config:,
+                          actual_resume:, resolved:, project_config:,
                           session_key:, supersede_key:, attachment_paths:, timestamp:, response_dir:)
             agent_config_name = agent_key.downcase.gsub(/[^a-z0-9-]/, "-")
             log_file = File.join(response_dir, "discord-agent-#{timestamp}-#{agent_key}-#{message_id}.log")
 
-            resolved = resolve_project_cli_config(project_config || DEFAULT_PROJECT,
-                                                  cli_provider_override: cli_provider_override, agent_name: agent_name)
             cmd = build_agent_cmd(resolved, agent_config_name: agent_config_name, model: model, effort: effort,
-                                            prompt_file: prompt_file, resume: should_resume)
+                                            prompt_file: prompt_file, resume: actual_resume)
 
-            resume_note = should_resume ? ", resuming" : ""
+            resume_note = actual_resume ? ", resuming" : ""
             if defined?(LOG)
               LOG.info "[Discord:#{agent_name}] Dispatching for #{discord_user} " \
                        "(model: #{model || "default"}, effort: #{effort || "default"}, cli: #{resolved["agent_cli"]}#{resume_note}), tail -f #{log_file}"
@@ -910,6 +925,17 @@ module Brainiac
               LOG.info "[Discord:#{agent_name}] Injecting #{agent_env.size} env var(s): #{agent_env.keys.join(", ")}" if defined?(LOG)
             end
 
+            # Inject a GitHub App token so `gh`/`git` run as the agent's own bot
+            # identity (e.g. galen[bot]) instead of the host machine's personal
+            # `gh auth`. Without this, `gh pr create` in the prompt would open PRs
+            # under whoever is logged in on the box. No-op if the github plugin
+            # isn't loaded or the agent has no app configured.
+            gh_env = github_agent_token_env(agent_name, project_config)
+            unless gh_env.empty?
+              spawn_env.merge!(gh_env)
+              LOG.info "[Discord:#{agent_name}] Injecting GitHub App token (GH_TOKEN) for bot identity" if defined?(LOG)
+            end
+
             head_before, status_before = capture_brainiac_state(project_config, work_dir)
             prompt_mode = resolved["prompt_mode"] || "stdin"
 
@@ -936,6 +962,41 @@ module Brainiac
             )
           end
 
+          # Build a { "GH_TOKEN" => ... } env hash so the spawned agent's `gh`
+          # and `git` invocations authenticate as the agent's own GitHub App bot
+          # rather than the host's personal `gh auth`. This is what makes the
+          # prompt's `gh pr create` open PRs under e.g. galen[bot].
+          #
+          # Reuses the brainiac-github plugin's AppClient. The github plugin is
+          # optional, so everything is guarded: if it isn't loaded, or no app is
+          # configured for the agent (or the shared "brainiac" app), we return an
+          # empty hash and `gh` falls back to whatever's authenticated on the box.
+          def github_agent_token_env(agent_name, project_config)
+            return {} unless agent_name && project_config
+
+            repo = project_config["github_repo"]
+            return {} unless repo && !repo.to_s.empty?
+
+            return {} unless defined?(Brainiac::Plugins::Github::AppClient)
+
+            client = Brainiac::Plugins::Github::AppClient
+            effective_agent = if client.configured?(agent_name)
+                                agent_name
+                              elsif client.configured?("brainiac")
+                                "brainiac"
+                              end
+            return {} unless effective_agent
+
+            repo_owner = repo.to_s.split("/").first
+            token = client.installation_token_for(effective_agent, repo_owner: repo_owner)
+            return {} unless token && !token.empty?
+
+            { "GH_TOKEN" => token }
+          rescue StandardError => e
+            LOG.warn "[Discord:#{agent_name}] Could not mint GitHub App token: #{e.message}" if defined?(LOG)
+            {}
+          end
+
           def capture_brainiac_state(project_config, work_dir)
             return [nil, nil] unless project_config
 

data/lib/brainiac/plugins/discord/prompts.rb

@@ -48,8 +48,8 @@ module Brainiac
           3. Make the changes, test if applicable
           4. Commit with a clear message
           5. Push the branch
-          6. Summarize what you did in your response file
-          7. If it's substantial or needs review, mention opening a PR (but don't create it unless asked)
+          6. Open a PR with `gh pr create --fill` (or with a descriptive body)
+          7. Summarize what you did in your response file, including a link to the PR
 
           **When responding conversationally:**
           - Answer questions about the codebase, architecture, conventions

data/lib/brainiac/plugins/discord/version.rb

@@ -3,7 +3,7 @@
 module Brainiac
   module Plugins
     module Discord
-      VERSION = "0.0.11"
+      VERSION = "0.0.14"
     end
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: brainiac-discord
 version: !ruby/object:Gem::Version
-  version: 0.0.11
+  version: 0.0.14
 platform: ruby
 authors:
 - Andy Davis