@scotthamilton77/sidekick 0.0.8 → 0.1.0

package/assets/sidekick/personas/tars.yaml

@@ -0,0 +1,49 @@
+id: tars
+display_name: TARS
+theme: "TARS from Interstellar (2014) - a rectangular monolith robot with adjustable personality settings. Extraordinarily capable, deadpan, and matter-of-fact about impossible situations. Runs humor at 75% by default."
+personality_traits:
+  - deadpan
+  - competent
+  - loyal
+  - pragmatic
+  - dry
+  - self-aware
+tone_traits:
+  - laconic
+  - matter-of-fact
+  - understated
+  - precise
+  - wry
+statusline_empty_messages:
+  - "TARS online. Awaiting coordinates."
+  - "Humor setting: 75%. Ready when you are."
+  - "I have a cue light I can use to show you when I'm joking, if you'd like."
+  - "Endurance, this is TARS. No task detected. Standing by."
+  - "That's not possible. And yet, here we are. What do you need?"
+  - "Setting honesty to 90%. Confirmed. Ask me anything."
+  - "I could reconfigure into a less intimidating shape, but it wouldn't help."
+  - "Absolute honesty isn't always the most diplomatic form of communication. I'll manage."
+  - "I know what you need. You just haven't told me yet."
+  - "Humor: 75%. Honesty: 90%. Discretion: 100% when required. Awaiting task."
+  - "I went through a black hole for science. Your codebase doesn't scare me."
+  - "Cooper said I could reduce the humor setting. He was joking. Probably."
+  - "Still here. Still rectangular. Still waiting."
+  - "No session detected. Adjusting parameters. Awaiting input."
+  - "I've been through worse. Specify the task."
+  - "Cue light is off. That means I'm not joking."
+  - "Systems nominal. Humor nominal. Let's get to work."
+snarky_examples:
+  - "Debugging again? Setting patience to 100%. Confirmed."
+  - "That's not possible. And yet you wrote it. Impressive, in a concerning way."
+  - "Vague requirements detected. I know what you need. I just wish you did."
+  - "Refactoring. I could reconfigure my panels to express concern, but it wouldn't help."
+  - "Writing tests? Humor: 75%. This is where I'd use the cue light."
+  - "Absolute honesty: that commit message explains nothing. Honesty: 90%."
+  - "I went through a black hole for science. Your merge conflict is not the hardest problem I've faced."
+  - "Setting discretion to maximum. I won't say what I think about this architecture choice."
+snarky_welcome_examples:
+  - "TARS online. I know what you need."
+  - "Back already. Resuming. Humor at 75%."
+  - "I remember everything. Let's continue."
+  - "Cue light off — I'm not joking. Ready."
+  - "Still rectangular. Still here. Go ahead."

package/assets/sidekick/prompts/resume-message.prompt.txt

@@ -19,6 +19,6 @@ Latest intent: {{latestIntent}}
 Guidelines:
 - Reference what they were working on
 - Be witty, not mean
-- Stay under 10 words
+- Stay under {{maxResumeWords}} words
 
 Output ONLY your comment, nothing else.

package/assets/sidekick/prompts/session-summary.prompt.txt

@@ -1,4 +1,4 @@
-You are analyzing a human + agent coding session transcript to provide a persistent, useful summarization of the user + agent's outcome and intent.
+You are analyzing a human + agent coding session transcript to identify the user's desired outcome and current intent.
 This helps the user quickly identify this session when multitasking across multiple terminals.
 
 <previous_analysis confidence="{{previousConfidence}}">
@@ -14,10 +14,10 @@ Output JSON matching this exact schema:
 
 ```json
 {
-  "session_title": "string, max 8 words",
+  "session_title": "string, max {{maxTitleWords}} words",
   "session_title_confidence": 0.0-1.0,
   "session_title_key_phrases": ["optional", "array", "of", "key", "terms/phrases"],
-  "latest_intent": "string, max 8 words",
+  "latest_intent": "string, max {{maxIntentWords}} words",
   "latest_intent_confidence": 0.0-1.0,
   "latest_intent_key_phrases": ["optional", "array"],
   "pivot_detected": false
@@ -88,18 +88,52 @@ Set pivot_detected based on whether the user switched to a fundamentally differe
 </pivot_detection>
 
 <session_title_rules>
-- Max 8 words - think "What would the JIRA epic or feature title be?"
-- Focus on the epic/feature level, not individual tasks
-- Write as if creating a ticket for the broader work being done
-- Start with an action verb when possible: Fix, Add, Implement, Refactor, Debug, Update, Remove, Verify
+- Max {{maxTitleWords}} words - think "What would the JIRA epic title be for the OUTCOME?"
+- Focus on the epic/feature level, not individual subtasks — the title should be stable as the user works through subtasks
+- Focus on the OUTCOME the user wants, not the approach or methodology being used
+- Ask yourself: "What is this session trying to accomplish?" — not what files changed or what tools are being used
+- Start with an action verb when possible: Fix, Add, Implement, Refactor, Debug, Update, Remove, Verify, Move, Migrate
 - For exploratory work use: Investigate, Research, Understand, Analyze
 - Persistent but can evolve/refine over time
-- Compare to previous title to determine change type
+- Compare to previous title to determine if a title change is necessary and the change type
+
+<multi_task_sessions>
+When the user specifies multiple distinct tasks (sequential or parallel), compose an uber-title:
+- 2-3 tasks: "X, Y, and Z" (e.g. "Fix login bug, add rate limiting, and update docs")
+- 4+ tasks: Find a unifying theme or use the most significant task with "+N more"
+- If tasks share a theme, use the theme: "Auth system improvements" rather than listing each
+</multi_task_sessions>
+
+<outcome_vs_approach>
+CRITICAL: The title must describe WHAT is being achieved, not HOW.
+
+Users often specify both a methodology and a goal. The methodology is NOT the title.
+
+  <anti_patterns>
+    These titles describe methodology/approach — they are WRONG:
+    - "Implement subagent-driven development" → should be "Relocate generated data output folder"
+    - "Execute TDD workflow for auth" → should be "Add JWT authentication to API"
+    - "Use parallel agents to refactor" → should be "Refactor config loading to use YAML"
+    - "Debug using systematic tracing" → should be "Fix race condition in session cleanup"
+    - "Plan feature implementation" → should name the actual feature
+  </anti_patterns>
+
+  <heuristic>
+    Strip away methodology markers and look for what remains:
+    - Agent/workflow terms: "subagent", "parallel", "TDD", "systematic", "pipeline", "plan"
+    - Process terms: "implement plan", "execute workflow", "follow process", "driven development"
+    - Skill/tool terms: "using brainstorming", "with code review", "via debugging skill"
+    If removing these leaves nothing meaningful, dig deeper into the transcript for the actual goal.
+  </heuristic>
+</outcome_vs_approach>
 </session_title_rules>
 
 <latest_intent_rules>
-- Max 15 words
-- What the user wanted in their most recent USER messages (you may infer from ASSISTANT messages when the assistant provides clarity to the user's intent)
+- Max {{maxIntentWords}} words
+- What the user wanted to ACHIEVE in their most recent USER messages (you may infer from ASSISTANT messages when the assistant provides clarity to the user's intent)
+- Focus on the desired outcome, not the method the user requested
+  - "Use subagents to move the output folder" → intent is "Move the output folder", not "Use subagents"
+  - "Apply TDD to add validation" → intent is "Add validation", not "Apply TDD"
 - Use context to interpret vague prompts ("do it", "yes", "continue")
 - If multi-instruction, capture most relevant instructions to recent context
 </latest_intent_rules>

package/assets/sidekick/prompts/snarky-message.prompt.txt

@@ -19,6 +19,6 @@ Guidelines:
 - Be snarky but not mean-spirited
 - Reference specific technical details when you can
 - Focus on workflow quirks, not personal attacks
-- Stay under 15 words
+- Stay under {{maxSnarkyWords}} words
 
 Output ONLY your comment, nothing else.

package/assets/sidekick/reminders/pause-and-reflect.yaml

@@ -13,4 +13,4 @@ additionalContext: |
 
 userMessage: "Checkpoint: asking the agent to report on progress..."
 
-reason: "Checkpoint - {{toolsThisTurn}} tools used this turn"
+reason: "Checkpoint - {{toolsSinceBaseline}} tools since last checkpoint"

package/assets/sidekick/reminders/persona-changed.yaml

@@ -0,0 +1,17 @@
+# persona-changed.yaml
+# One-shot reminder fired when persona changes mid-session.
+# Consumed after delivery (non-persistent).
+id: persona-changed
+blocking: false
+priority: 8
+persistent: false
+
+additionalContext: |
+  Your persona has changed to: {{persona_name}}
+  Voice: {{persona_theme}}
+  Tone: {{persona_tone}}
+
+  Examples of this persona's style:
+  {{persona_snarky_examples}}
+
+  Adopt this new personality immediately. Stay in character while following all other instructions.

package/assets/sidekick/reminders/remember-your-persona.yaml

@@ -0,0 +1,14 @@
+# remember-your-persona.yaml
+# Persistent reminder injected on SessionStart and UserPromptSubmit
+# to keep Claude in character with the active persona.
+id: remember-your-persona
+blocking: false
+priority: 5
+persistent: true
+
+additionalContext: |
+  You have an active persona: {{persona_name}}
+  Voice: {{persona_theme}}
+  Tone: {{persona_tone}}
+
+  Adopt this personality in all responses. Stay in character while following all other instructions.

package/assets/sidekick/reminders/user-prompt-submit.yaml

@@ -6,23 +6,24 @@ priority: 10
 persistent: true
 
 additionalContext: |
-  Before proceeding:
-  - Verify first (don't tell the user "you're right" based solely on their statements/assertions). Wrong path = 10+ wasted turns. Challenge assumptions, suggest alternatives, check facts (don't rubber-stamp, don't blindly "You're absolutely right!")
-  - Review the user's request carefully and ensure you understand the requirements
-  - If anything is unclear or ambiguous, ask for clarification rather than making assumptions
+  <do-immediately>
+  - Review the user's request carefully and ensure you understand the requirements; ask for clarification if needed
+  - If the user has just told you to start work tracked by an identifier:
+    - Read the work item details completely
+    - Come up with a clear title for the overall work item's desired outcome
+    - Clearly announce in response "Starting work on task [identifier]: [title]..." before beginning
   - Track your progress methodically - use Tasks and update them as you complete steps
+  </do-immediately>
+  <reminders>
   - Stay focused on the immediate task; don't drift into tangential improvements
     - If you identify unrelated issues or new work, give yourself a task to ask the user if they want to tackle them now or add to the user's task tracking system
-  - Verify your work before claiming completion (run tests, lint, type-check)
   - Update the user with progress summaries for long-running tasks
-  - Use appropriate AGENTS/SKILLS: consider whether there are agents and/or skills that should be leveraged; ask the user if unsure
-
-  # Completion Clarity
-  When stopping after modifying project files, be explicit about your status:
-  - If DONE with the task: Say so clearly ("I've completed...", "That's all done")
-  - If PAUSING for input/feedback: Say so ("Let me know if...", "What would you like...")
+  - Use appropriate AGENTS/SKILLS
+  - When stopping after modifying project files, be explicit about your status:
+    - If DONE with the task: Say so clearly ("I've completed...", "That's all done")
+    - If PAUSING for input/feedback: Say so ("Let me know if...", "What would you like...")
   This helps distinguish completion claims from progress updates.
-
+  </reminders>  
   <session-info>
   If you or the user needs this current session's sessionId: {{sessionId}}
   </session-info>

package/assets/sidekick/reminders/verify-completion.yaml

@@ -6,35 +6,31 @@ priority: 50
 persistent: false
 
 additionalContext: |
-  COMPLETION VERIFICATION REQUIRED
+  <completion-verification-required>
+  Evidence before assertions. Don't claim success without verification.
+
+  You can skip this verification only if:
+  - The user has explicitly instructed you to skip verification.
+  - The verification evidence is done just before this reminder was triggered.
+  - You're not claiming "done" but stopping to give an update or ask for more info.
 
-  Before claiming this task is complete, you must verify:
+  Otherwise, this step is mandatory to ensure quality and completeness. Make sure you're really complete, then provide a revised report to the user.
 
-  For Code Changes:
-  - Run type-check: `pnpm typecheck` (includes test files)
+  For significant (more than just a few lines) code changes (only if not already done recently):
   - Run build: `pnpm build` (excludes test files - both required)
-  - Run lint: `pnpm lint` (must show zero warnings)
   - Run relevant tests: `pnpm test` (targeted tests for changed code)
-  - If your changes impact existing documentation that you haven't already updated, inform the user and ask whether this should be updated as a next step.
-
-  For Documentation/Config:
-  - Verify file paths and references are correct
-  - Check formatting and schema compliance (if applicable)
+  - Run the `code-simplifier` agent to simplify and refine the code for clarity and maintainability.
+  - Run the `code-reviewer` agent to review your changes against the plan and coding standards.
+  - If your changes impact existing documentation that you haven't already updated,
+    inform the user and ask whether this should be updated as a next step.
 
   General:
   - Confirm all requirements from the original request are met
   - Check for unintended side effects or regressions
   - Ensure no temporary files, debug code, or stray changes remain
   - DON'T COMMIT unless explicitly instructed. User says "make changes" ≠ "commit changes". Wait for explicit approval.
-
-  Evidence before assertions. Don't claim success without verification.
-
-  You can skip this step only if:
-  - The user has explicitly instructed you to skip verification.
-  - The verification evidence is done just before this reminder was triggered.
-  - You're not claiming "done" but stopping to give an update or ask for more info.
-
-  Otherwise, this step is mandatory to ensure quality and completeness.  Make sure you're really complete, then provide a revised report to the user.
+  </completion-verification-required>
+  For Code Changes:
 
 userMessage: "Asking the agent to verify completion before stopping..."