qualia-framework 2.6.0 → 3.1.0

package/framework/skills/qualia-evolve/SKILL.md

@@ -1,200 +0,0 @@
----
-name: qualia-evolve
-description: "Self-optimization loop — analyzes metrics, Lab Notes, and session history across all shipped projects, then proposes framework improvements. The Karpathy loop for the framework itself. Use after shipping a project, or periodically to improve the framework."
-user-invocable: true
-args:
-  - name: project
-    description: "Project to collect metrics from (defaults to current directory)"
-    required: false
----
-
-# Qualia Evolve — Framework Self-Optimization
-
-The recursive improvement loop. Reads real project outcomes, identifies what's working and what's failing, and proposes targeted changes to skill prompts, agent instructions, templates, and defaults.
-
-**This is NOT code optimization.** This optimizes the FRAMEWORK — the prompts, the flow, the defaults, the instructions that shape how every future project runs.
-
-## Usage
-
-`/qualia-evolve` — Full optimization cycle (collect + analyze + propose)
-`/qualia-evolve --collect` — Just collect metrics from current project
-`/qualia-evolve --analyze` — Just analyze existing metrics (no collection)
-
-## Process
-
-### 1. Collect Metrics (if not --analyze only)
-
-Run the metrics harness on the current project:
-
-```bash
-bash ~/.claude/qualia-framework/bin/collect-metrics.sh "$(pwd)"
-```
-
-This appends a row to `~/.claude/knowledge/framework-metrics.md`.
-
-### 2. Load All Evidence
-
-Read these files — they're the raw data for optimization:
-
-```bash
-# Framework metrics (all shipped projects)
-cat ~/.claude/knowledge/framework-metrics.md 2>/dev/null
-
-# Session digest (recent sessions across all projects)
-cat ~/.claude/knowledge/session-digest.md 2>/dev/null
-
-# Global learned patterns
-cat ~/.claude/knowledge/learned-patterns.md 2>/dev/null
-
-# Optimization objective (what "better" means)
-cat ~/.claude/qualia-framework/OBJECTIVE.md
-```
-
-Also scan for Lab Notes across all projects:
-
-```bash
-find ~/Projects/*/. -name "LAB-NOTES.md" -path "*/.planning/*" 2>/dev/null | while read f; do
-  echo "=== $(dirname $(dirname $f) | xargs basename) ==="
-  cat "$f"
-done
-```
-
-### 3. Spawn Analysis Agents (Fan-out)
-
-Spawn 3 parallel agents to analyze different dimensions:
-
-**Agent 1: Flow Efficiency Analyst**
-```
-Analyze framework metrics and session digests.
-- Which projects took the most sessions? Why?
-- Where do employees get stuck? (look for /qualia-idk calls, long sessions, repeated phases)
-- Which phases consistently need gap-fix plans? (high deviation signal)
-- Are preferences defaults working? (do people customize, or use recommended?)
-- What's the trend? Is the framework getting faster or slower?
-Return: top 3 flow bottlenecks with evidence.
-```
-
-**Agent 2: Failure Pattern Analyst**
-```
-Analyze Lab Notes across all projects and learned-patterns.md.
-- What approaches keep failing? Are we still recommending them in skill prompts?
-- Are there common anti-patterns that multiple projects hit?
-- Which agent instructions lead to the most deviations?
-- Are there failure categories? (e.g., "auth always needs 2 tries", "Supabase RLS is always missed first pass")
-Return: top 3 recurring failure patterns with root cause.
-```
-
-**Agent 3: Quality Gate Analyst**
-```
-Analyze verification pass rates and post-deploy issues.
-- Which phases fail verification most often? What type of failures?
-- Are the quality gates catching real issues or adding overhead?
-- Is the red-team QA agent finding things the verifier misses?
-- What gets caught at /qualia-review that should have been caught earlier?
-Return: top 3 quality gaps with evidence.
-```
-
-### 4. Synthesize Findings
-
-Spawn the synthesizer agent with all 3 reports:
-
-```
-Downstream consumer: Fawzi reviewing framework improvement proposals.
-Cross-reference findings. Identify root causes that explain multiple symptoms.
-A flow bottleneck might be caused by a failure pattern which is caused by a quality gap.
-```
-
-### 5. Propose Framework Changes
-
-For each finding, propose a SPECIFIC change:
-
-```markdown
-## Proposed Changes
-
-### Change 1: [title]
-**Evidence:** [metrics/lab notes that support this]
-**Root cause:** [why this is happening]
-**Proposed fix:** [exact file + exact change]
-**Expected impact:** [which metric improves and by how much]
-**Risk:** [what could go wrong]
-
-### Change 2: ...
-```
-
-Rules for proposals:
-- Every change must cite evidence from metrics or Lab Notes
-- Every change must target a specific file (skill, agent, template, or default)
-- Every change must predict which metric it improves
-- Maximum 5 changes per cycle (don't over-optimize)
-- Never change OBJECTIVE.md or collect-metrics.sh (those are the fixed harness)
-
-### 6. Present to Fawzi for Approval
-
-Show the proposals with evidence. For each one:
-
-```
-"Change [N]: [title]
- Evidence: [data]
- File: [path]
- Change: [what specifically changes]
-
- Apply? (yes/no/modify)"
-```
-
-### 7. Apply Approved Changes
-
-For each approved change:
-1. Read the target file
-2. Make the specific edit
-3. Sync to `~/Projects/qualia-framework/` (the git-tracked copy)
-4. Log the change to `~/.claude/knowledge/framework-evolution.md`
-
-### 8. Commit Evolution
-
-```bash
-cd ~/Projects/qualia-framework && git add -A && git commit -m "evolve: [summary of changes]"
-```
-
-## Evolution Log
-
-Every change gets logged to `~/.claude/knowledge/framework-evolution.md`:
-
-```markdown
-## [Date] — Cycle [N]
-
-**Trigger:** Post-ship analysis of [project]
-**FQS before:** [score]
-
-### Changes Applied
-1. [change description] → [file modified]
-2. ...
-
-### Changes Rejected
-1. [change description] — Reason: [why]
-
-### Predicted Impact
-- [metric]: [expected change]
-```
-
-This log is read in future cycles to track whether changes actually helped.
-
-## Bounded Time Budget
-
-The optimization pass should complete in ~10 minutes:
-- Metrics collection: 30 seconds
-- Evidence loading: 1 minute
-- 3 analysis agents (parallel): 3-5 minutes
-- Synthesis: 2 minutes
-- Proposal + approval: 2-3 minutes
-
-Do NOT spend more than 15 minutes total. If analysis is taking too long, present what you have.
-
-## Agents Used
-
-| Agent | Role |
-|-------|------|
-| `general-purpose` (x3) | Flow, Failure, Quality analysis |
-| `qualia-research-synthesizer` | Cross-cutting synthesis |
-
----
-> Run `/qualia-evolve` after shipping a project. The framework gets smarter every time.

package/framework/skills/qualia-execute-phase/SKILL.md

@@ -1,89 +0,0 @@
----
-name: qualia-execute-phase
-description: "Wave-based parallel execution of phase plans with subagent orchestration. Use this skill whenever the user says 'execute phase', 'run phase', 'build phase', 'start building', 'execute', or wants to implement a planned phase. Also trigger when user mentions 'start implementation', 'begin coding', or has a planned phase ready to execute. Supports gap-closure mode for fixing issues found during verification."
----
-
-# Qualia Execute Phase — Phase Execution
-
-Execute all plans in a phase using wave-based parallel subagent orchestration.
-
-## Usage
-
-`/qualia-execute-phase 3` — Execute phase 3
-`/qualia-execute-phase` — Auto-detect next phase ready for execution
-`/qualia-execute-phase 3 --gaps-only` — Execute only gap closure plans (after verify-work found issues)
-
-## Process
-
-### 1. Load Plans
-
-Read all PLAN.md files for the target phase from `.planning/phases/{phase}-{slug}/`.
-
-For `--gaps-only`, read only the gap-closure plans created by `plan-phase --gaps`.
-
-Reference: `~/.claude/qualia-framework/workflows/execute-phase.md`
-UI/brand reference: `~/.claude/qualia-framework/references/ui-brand.md`
-
-### 2. Organize into Waves
-
-Group plan steps into execution waves based on dependencies:
-- **Wave 1**: Steps with no dependencies (can run in parallel)
-- **Wave 2**: Steps that depend on Wave 1 outputs
-- **Wave 3**: Steps that depend on Wave 2 outputs
-- etc.
-
-### 3. Execute Waves
-
-For each wave, spawn subagents in parallel. The orchestrator uses ~15% of context for coordination; each subagent gets 100% fresh context with:
-- The specific plan step(s) to execute
-- Project context (PROJECT.md, relevant files)
-- UI/brand guidelines
-- Clear acceptance criteria
-
-### 4. Collect Results
-
-As subagents complete:
-- Verify each step's acceptance criteria are met
-- Resolve any conflicts between parallel changes
-- Handle failures (retry or escalate to user)
-
-### 5. Create Phase Summary
-
-Write `{phase}-SUMMARY.md`:
-```markdown
-# Phase {N}: {Name} — Execution Summary
-
-## Completed Steps
-- Step 1: [what was done]
-- Step 2: [what was done]
-
-## Files Created/Modified
-- file1.tsx — [description]
-- file2.ts — [description]
-
-## Notes
-- Any deviations from plan
-- Tech debt incurred
-- Things to watch for in verification
-```
-
-### 6. Update State
-
-Update `.planning/STATE.md` phase status to `executed`.
-
-### 7. Next Step
-
-> "Phase {N} executed. Run `/qualia-verify-work {N}` to test the results."
-
-### Agents Used
-| Agent | File | Role |
-|-------|------|------|
-| `qualia-executor` | `~/.claude/agents/qualia-executor.md` | Executes each plan's tasks atomically |
-| `qualia-verifier` | `~/.claude/agents/qualia-verifier.md` | Verifies phase goal achievement |
-| `qualia-planner` | `~/.claude/agents/qualia-planner.md` | Creates gap closure plans (if verify fails) |
-| `qualia-plan-checker` | `~/.claude/agents/qualia-plan-checker.md` | Validates gap closure plans |
-
-Domain skill context from PLAN.md `@skill` references is automatically resolved and inlined into each executor subagent prompt. Plans referencing skills like `frontend-design`, `supabase`, `financial-ledger`, etc. will have those patterns available to executors.
-
----
-> Stuck? Type `/qualia-idk` · Lost? Type `/qualia-help` · Done for today? Type `/qualia-report`