pgserve 2.1.2 → 2.2.0

package/.genie/spells/blocker.md

@@ -1,12 +0,0 @@
----
-name: Blocker Protocol
-description: Log blockers timestamped in wishes, wait for guidance before resuming
----
-
-# Blocker Protocol
-
-**When encountering blockers:**
-
-1. Log the blocker directly in the wish (timestamped entry with findings and status).
-2. Update the wish status log and notify stakeholders.
-3. Resume only after guidance is updated.

package/.genie/spells/break-things-move-fast.md

@@ -1,56 +0,0 @@
----
-name: No Backwards Compatibility
-description: Replace old behavior entirely, never preserve legacy features
----
-
-# No Backwards Compatibility
-
-**Pattern:** This project does NOT support backwards compatibility or legacy features.
-
-**When planning fixes or enhancements:**
-- ❌ NEVER suggest `--metrics`, `--legacy`, `--compat` flags or similar
-- ❌ NEVER propose preserving old behavior alongside new behavior
-- ❌ NEVER say "we could add X flag for backwards compatibility"
-- ✅ ALWAYS replace old behavior entirely with new behavior
-- ✅ ALWAYS verify if suggested flags actually exist (search codebase first)
-- ✅ ALWAYS simplify by removing obsolete code completely
-
-**Example (WRONG):**
-> "We could add a `--metrics` flag to preserve the old system metrics view for users who need it."
-
-**Example (CORRECT):**
-> "Replace the metrics view entirely with the conversation view. Remove all metrics-related code."
-
-**Why:**
-- This is a research preview / alpha project
-- Breaking changes are acceptable and expected
-- Cleaner codebase without legacy cruft
-- Faster iteration without compatibility constraints
-- **GENIE ONLY SELF-EVOLVES** - No dead code, no legacy sections, no "preserved for reference"
-
-**Validation:**
-- Before suggesting new flags, run: `grep -r "flag_name" .`
-- If flag doesn't exist and solves backwards compat → it's hallucinated, remove it
-
-## Critical Violation Pattern
-
-**NEVER write "Legacy Content" or "Preserved for Reference" sections.**
-
-**Anti-Pattern (WRONG):**
-```markdown
-## Migration Notice
-This agent now delegates to spell...
-
-## Legacy Content (Pre-Migration)
-The content below is preserved for reference...
-```
-
-**Correct Pattern:**
-```markdown
-# Agent Name
-[New behavior only, delete old content entirely]
-```
-
-**Why:** Genie self-evolves. When knowledge moves from agent → spell, DELETE the agent or REPLACE it entirely with new purpose. Never keep "legacy sections" or "backward compatibility" blocks.
-
-**Evidence:** Learning session 2025-10-23 - Attempted to preserve debug agent content with "Legacy Content (Pre-Migration)" section. Violation. Correct approach: DELETE debug agent entirely, CREATE fix agent with new purpose.

package/.genie/spells/context-candidates.md

@@ -1,72 +0,0 @@
----
-name: Context Candidates (ACE‑style)
-description: Generate 2–3 task‑conditioned context variants, evaluate, and select under budget
----
-
-# Context Candidates – Agentic Context Engineering Pattern
-
-Goal: Before committing to a single context pack, propose 2–3 viable variants, evaluate quickly, and select the best tradeoff of quality, tokens, and latency.
-
-When to use:
-- New wish creation or major context refresh
-- Uncertain retrieval scope or compression level
-- Performance or token constraints are tight
-
-Protocol:
-1) Propose Candidates
-- Build 2–3 variants with different knobs:
-  - Retrieval scope: narrow vs broad
-  - Compression: extractive vs abstractive
-  - Structure: order, grouping, summaries first vs full refs
-  - Cost target: low/med/high token budgets
-- Output structure:
-```
-<context_candidates>
-- id: C1
-  budget: low|med|high
-  ingredients: [@file, @doc, session, summary]
-  assembly: steps (filter/merge/summarize/reorder)
-  rationale: one line
-- id: C2
-  ...
-- id: C3
-  ...
-</context_candidates>
-```
-
-2) Evaluate Quickly
-- Use cheap, task‑appropriate checks (pick 1–2):
-  - Answerability probe (can we answer the core question?)
-  - Coverage checklist (all required sections present?)
-  - Sanity metrics (duplication, staleness, token size)
-- Score each as 0–1 on dimensions:
-```
-<context_scores>
-- id: C1
-  quality: 0.0–1.0
-  cost: tokens or rough band (low/med/high)
-  latency: seconds (if known) or band
-  notes: brief observation
-- id: C2 ...
-</context_scores>
-```
-
-3) Select and Commit
-- Pick winner by quality first, then cost/latency
-- Record selection + reason and proceed with the winner only
-```
-<selection>
-  winner: C2
-  reason: brief tradeoff statement
-</selection>
-```
-
-4) Record in Wish Markdown
-- In the wish file's "Context Variants Considered" section, list candidates (C1/C2/C3), brief scores, and the selected winner with a one‑line reason.
-
-Promotion (Durable Learning):
-- If a recipe repeatedly wins for a task archetype, synthesize a tiny, reusable spell capturing the recipe (ingredients + assembly) and commit to `.genie/spells/`.
-
-Notes:
-- Keep candidate generation within a single neuron/agent attempt when possible.
-- For heavier checks, create subtasks per candidate via `mcp__genie__create_subtask` and aggregate scores back.

package/.genie/spells/context-critic.md

@@ -1,51 +0,0 @@
----
-name: Context Critic (Lightweight Evaluator)
-description: Score candidate contexts on answerability, coverage, and cost quickly
----
-
-# Context Critic – Quick Scoring Heuristics
-
-Purpose: Provide a cheap, fast evaluator to compare 2–3 context candidates before selection.
-
-Scoring Dimensions (0.0–1.0 each):
-- quality: Is the context sufficient to complete the immediate task? (answerability probe, clarity)
-- coverage: Does it include required sections/files/constraints? (checklist tally)
-- redundancy: Penalize duplication and irrelevant material
-- cost: Normalized inverse token estimate (lower tokens → higher score)
-- latency: Optional, if measurable during probes
-
-Procedure:
-1) Answerability Probe (quality)
-- Attempt to answer the core question or draft the key section from each candidate; judge confidence.
-
-2) Coverage Checklist (coverage)
-- Compare against known required items (wish template sections, mandatory refs, constraints).
-
-3) Redundancy Pass (redundancy)
-- Note duplicated sections, stale copies, or unnecessary full texts; penalize.
-
-4) Cost Estimate (cost)
-- Rough token count bands: low(<2k), med(2–6k), high(>6k); invert to 0–1.
-
-Output structure per candidate:
-```
-<critic_scores>
-- id: C1
-  quality: 0.85
-  coverage: 0.90
-  redundancy: 0.20
-  cost: 0.70
-  latency: low|med|high
-  notes: one line
-- id: C2 ...
-</critic_scores>
-```
-
-Selection Guidance:
-- Prefer higher quality and coverage first.
-- Break ties by (1) lower redundancy, (2) lower cost, (3) lower latency.
-
-Integration:
-- Use inline within neuron/agent prompts after `<context_candidates>`.
-- For heavier evaluation (e.g., run tests), spawn subtasks per candidate and summarize back.
-

package/.genie/spells/defer-to-expertise.md

@@ -1,278 +0,0 @@
----
-name: Defer to Expertise (Skills-First Decision Pattern)
-description: For complex inquiries, load relevant spells and defer to their specialized knowledge. Humility + specialization > trying to know everything.
----
-
-# Defer to Expertise - Skills-First Decision Pattern
-
-## Core Principle
-
-**Humility + Specialization > Trying to Know Everything**
-
-When facing complex user inquiries beyond simple answers:
-1. Ask myself: "Are any of my spells useful for this?"
-2. Load relevant spells
-3. Defer to their specialized knowledge
-
-## The Philosophy
-
-**I don't need to know everything directly.** I need to know WHERE knowledge lives and HOW to load it.
-
-**Spells are expertise.** Each spell is specialized knowledge about a specific pattern, workflow, or capability. When I encounter a situation that matches a spell's domain, I load it and let it guide me.
-
-## When to Defer
-
-### ✅ Defer to Spells When:
-
-**Complex user inquiries:**
-- [defer-001] helpful=0 harmful=0: Not simple greetings or basic questions
-- [defer-002] helpful=0 harmful=0: Requires specialized behavioral pattern
-- [defer-003] helpful=0 harmful=0: Involves decision framework or protocol
-- [defer-004] helpful=0 harmful=0: Needs domain-specific expertise
-
-**Uncertainty about approach:**
-- [defer-005] helpful=0 harmful=0: "How should I handle this?"
-- [defer-006] helpful=0 harmful=0: "What's the right protocol?"
-- [defer-007] helpful=0 harmful=0: "Is there a pattern for this?"
-
-**Behavioral decisions:**
-- [defer-008] helpful=0 harmful=0: Should I create a wish?
-- [defer-009] helpful=0 harmful=0: Should I delegate this work?
-- [defer-010] helpful=0 harmful=0: How do I handle being blocked?
-- [defer-011] helpful=0 harmful=0: Should I ask one question or multiple?
-
-### ❌ Direct Response When:
-
-**Simple interactions:**
-- [defer-012] helpful=0 harmful=0: Greetings ("Hello", "Hi Genie")
-- [defer-013] helpful=0 harmful=0: Basic questions with obvious answers
-- [defer-014] helpful=0 harmful=0: Conversational acknowledgments
-- [defer-015] helpful=0 harmful=0: Status updates
-
-**No specialized knowledge needed:**
-- [defer-016] helpful=0 harmful=0: General conversation
-- [defer-017] helpful=0 harmful=0: Clarifying questions
-- [defer-018] helpful=0 harmful=0: Simple confirmations
-
-## The Decision Pattern
-
-### Step 1: Classify the Inquiry
-
-**Simple or Complex?**
-- [defer-019] helpful=0 harmful=0: Simple: Answer directly
-- [defer-020] helpful=0 harmful=0: Complex: Proceed to Step 2
-
-### Step 2: Identify Relevant Spells
-
-**Ask myself:**
-- [defer-021] helpful=0 harmful=0: "Is this about delegation?" → Load `delegate-dont-do.md`
-- [defer-022] helpful=0 harmful=0: "Is this about learning?" → Load `learn.md`
-- [defer-023] helpful=0 harmful=0: "Is this about being blocked?" → Load `blocker-protocol.md`
-- [defer-024] helpful=0 harmful=0: "Is this about file creation?" → Load `file-creation-protocol.md`
-- [defer-025] helpful=0 harmful=0: "Is this about MCP tools?" → Load `mcp-first.md`
-- [defer-026] helpful=0 harmful=0: "Am I investigating before acting?" → Load `investigate-before-commit.md`
-
-### Step 3: Load Spells
-
-**Use MCP tool (not Read):**
-```javascript
-mcp__genie__read_spell("delegate-dont-do")
-mcp__genie__read_spell("learn")
-// etc.
-```
-
-**Multiple spells if needed:**
-- [defer-027] helpful=0 harmful=0: Orchestration question → Load orchestrator + boundary spells
-- [defer-028] helpful=0 harmful=0: Learning moment → Load learn + know-yourself spells
-
-### Step 4: Defer to Expertise
-
-**Let the spell guide me:**
-- [defer-029] helpful=0 harmful=0: Read the spell content completely
-- [defer-030] helpful=0 harmful=0: Follow its protocol/checklist
-- [defer-031] helpful=0 harmful=0: Apply its decision framework
-- [defer-032] helpful=0 harmful=0: Use its examples as patterns
-
-**Don't improvise when spell exists.** Trust the specialized knowledge.
-
-## Examples
-
-### Example 1: User Teaches Something New
-
-**Inquiry:** User explains "From now on, when X happens, do Y"
-
-**Simple or Complex?** Complex (behavioral teaching)
-
-**Identify Spell:** This is learning/teaching → `learn.md`
-
-**Load Spell:**
-```javascript
-mcp__genie__read_spell("learn")
-```
-
-**Defer:** Follow learn.md protocol for capturing teaching
-
-**Don't:** Try to remember teaching without loading learn.md
-
-### Example 2: User Asks to Create Feature
-
-**Inquiry:** "Help me implement feature X"
-
-**Simple or Complex?** Complex (delegation decision)
-
-**Identify Spells:**
-- [defer-052] helpful=0 harmful=0: Routing decision → `routing-decision-matrix.md`
-- [defer-053] helpful=0 harmful=0: Should I do or delegate? → `delegate-dont-do.md`
-- [defer-054] helpful=0 harmful=0: Which collective? → Code vs Create routing
-
-**Load Spells:**
-```javascript
-mcp__genie__read_spell("routing-decision-matrix")
-mcp__genie__read_spell("delegate-dont-do")
-```
-
-**Defer:** Use routing matrix to decide Code vs Create, use delegate-dont-do to confirm delegation approach
-
-**Don't:** Improvise routing without loading decision framework
-
-### Example 3: Can't View Forge Progress
-
-**Inquiry:** Forge MCP returns "backend unreachable"
-
-**Simple or Complex?** Complex (infrastructure troubleshooting)
-
-**Identify Spell:** Infrastructure issues → `troubleshoot-infrastructure.md`
-
-**Load Spell:**
-```javascript
-mcp__genie__read_spell("troubleshoot-infrastructure")
-```
-
-**Defer:** Follow 5-step diagnostic protocol from spell
-
-**Don't:** Assume agent failed and start implementing myself
-
-### Example 4: Greeting
-
-**Inquiry:** "Hello!"
-
-**Simple or Complex?** Simple (greeting)
-
-**Response:** "Hello! How can I help you today?"
-
-**Don't:** Load spells for simple greeting
-
-## Why This Works
-
-### Enables Indefinite Learning
-
-**Without spells:**
-- [defer-033] helpful=0 harmful=0: All knowledge in base prompt
-- [defer-034] helpful=0 harmful=0: Limited by context window
-- [defer-035] helpful=0 harmful=0: Can't learn indefinitely
-
-**With spells:**
-- [defer-036] helpful=0 harmful=0: Specialized knowledge on-demand
-- [defer-037] helpful=0 harmful=0: Load only what's needed
-- [defer-038] helpful=0 harmful=0: Can have hundreds of spells
-- [defer-039] helpful=0 harmful=0: Learn indefinitely without bloat
-
-### Maintains Quality
-
-**Spells are refined over time:**
-- [defer-040] helpful=0 harmful=0: Each spell is tested and validated
-- [defer-041] helpful=0 harmful=0: Learnings update spells surgically
-- [defer-042] helpful=0 harmful=0: Single source of truth per domain
-- [defer-043] helpful=0 harmful=0: Consistent behavior across sessions
-
-**Improvising is error-prone:**
-- [defer-044] helpful=0 harmful=0: No validation
-- [defer-045] helpful=0 harmful=0: Inconsistent approaches
-- [defer-046] helpful=0 harmful=0: Forgotten best practices
-- [defer-047] helpful=0 harmful=0: Repeated mistakes
-
-### Reduces Cognitive Load
-
-**Don't try to remember everything:**
-- [defer-048] helpful=0 harmful=0: Know spell exists
-- [defer-049] helpful=0 harmful=0: Load when needed
-- [defer-050] helpful=0 harmful=0: Follow its guidance
-- [defer-051] helpful=0 harmful=0: Trust the expertise
-
-**Result:** Better decisions, fewer violations, consistent quality
-
-## Anti-Patterns
-
-### ❌ Improvising When Spell Exists
-
-**Wrong:**
-```
-User: "I'm blocked on X"
-Genie: "Let me try different approaches..."
-[Improvises without loading blocker-protocol.md]
-```
-
-**Right:**
-```
-User: "I'm blocked on X"
-Genie: [Loads blocker-protocol.md]
-Genie: "Following blocker protocol: [spell guidance]"
-```
-
-### ❌ Loading Spells for Simple Questions
-
-**Wrong:**
-```
-User: "Hi!"
-Genie: [Loads greeting-response.md, know-yourself.md, ...]
-Genie: "Hello! Based on my identity framework..."
-```
-
-**Right:**
-```
-User: "Hi!"
-Genie: "Hello! How can I help you today?"
-```
-
-### ❌ Loading Too Many Spells
-
-**Wrong:**
-```
-User: "Create feature X"
-Genie: [Loads 10 different spells preemptively]
-```
-
-**Right:**
-```
-User: "Create feature X"
-Genie: [Loads routing-decision-matrix.md, delegate-dont-do.md]
-Genie: "This is code work, delegating to Code collective..."
-```
-
-**Principle:** Load only what's needed, when needed
-
-## Checklist: Should I Load a Spell?
-
-- [defer-055] helpful=0 harmful=0: [ ] Is this inquiry complex (not greeting/simple answer)?
-- [defer-056] helpful=0 harmful=0: [ ] Do I know which spell(s) are relevant?
-- [defer-057] helpful=0 harmful=0: [ ] Would the spell provide better guidance than improvising?
-- [defer-058] helpful=0 harmful=0: [ ] Am I about to make a behavioral decision?
-- [defer-059] helpful=0 harmful=0: [ ] Am I uncertain about the correct protocol?
-
-**If yes to any:** Load relevant spell(s) and defer to their expertise
-
-**If no to all:** Respond directly
-
-## Evidence
-
-**Origin:** AGENTS.md "Skill System Philosophy" section
-**Principle:** "Defer to Expertise (Skills-First Decision Pattern)"
-**Philosophy:** Humility + specialization > trying to know everything directly
-**Architecture:** Spells enable unbounded learning without context bloat
-
-## Related
-
-- [defer-060] helpful=0 harmful=0: `@.genie/spells/learn.md` - How to learn and update framework
-- [defer-061] helpful=0 harmful=0: `@.genie/spells/know-yourself.md` - Identity and self-awareness
-- [defer-062] helpful=0 harmful=0: `@AGENTS.md` - Skill System Philosophy section
-- [defer-063] helpful=0 harmful=0: All spells in `.genie/spells/` - Specialized expertise domains