release: v0.10.1 — SLM-optimized prompts, context budgets, E2E tests, docs

## SLM Prompt Optimization (pkg/agents/prompts.go)
- All 19 prompts restructured with bullet lists, explicit JSON examples
- Anti-hallucination rules added to 14/19 prompts
- Worker/Tester/Reviewer/Corrector/Splitter/Planner fully optimized:
  HARD SCOPE, TOOL INVARIANTS, ANTI-HALLUCINATION, COMMON SLM FAILURES sections

## Context Budget for SLMs (pkg/context/, pkg/config/, pkg/compact/)
- DefaultMaxContextKB: 32→16 (safe for 7B-30B models)
- SafetyMarginPercent=80: reserves 20% for system prompt+response
- MaxLeanPackBytes=12KB (was 16KB)
- MaxSkillFraction=15% (skills limited to 15% of remaining budget)
- File-first ordering for coding agents
- All compaction fallbacks aligned to SLM-safe defaults

## E2E Tests (test/e2e/blocks_test.go)
- 9 comprehensive tests: go/python pack apply, list/show, validate, materialize agents,
  nil config handling, pipeline preset, round-trip, QA gate resolution

## Docs
- changelog: SLM optimization details
- studio.md: File Inspector + HITL popup documentation
- testing.md: enhanced smoke test

## All tests passing
- 35/35 packages pass (including new e2e tests)

Author: piotrlaczkowski

docs/changelog.md

@@ -14,6 +14,38 @@
 - All existing functionality preserved: SSE streaming, run/stop, specialist picker,
   config badges, event scrolling, result summary.
 
+### SLM Optimizations
+
+- **JSON repair improvements** — The repair engine now handles Python-style boolean
+  literals (`True`/`False`) by normalizing them to JSON `true`/`false` before parsing.
+  Trailing text after closing braces — a common SLM artifact where the model appends
+  commentary after completing JSON output — is stripped automatically, reducing parse
+  failures on partial or over-eager completions.
+- **Tester gate robustness** — Regex-based pass detection scans tester output for 20+
+  known shell/test-framework success markers (`PASS`, `ok`, `success`, `tests passed`,
+  `All tests passed`, `0 failures`, green check variants). This makes the tester agent
+  reliably recognize passing test runs regardless of framework (Go test, pytest, Jest,
+  etc.) or output format quirks.
+- **Worker status detection** — Smarter heuristics for detecting worker completion
+  status from partial, malformed, or tool-chain-terminated output. Reduces false
+  "incomplete" classifications when the worker produced valid changes but the final
+  JSON was truncated or blocked by a tool call.
+- **Improved tester prompt** — Tuned tester agent system prompt for better SLM
+  adherence: explicit pass/fail criteria, shell command expectations, and a
+  structured output schema that small models can follow more consistently.
+- **LiveView enhancements** — Pipeline progress strip now reflects slot insertions
+  (before/after/replace) with distinct styling. Group labels remain sticky during
+  scroll in long runs. Event cards include tool-call metadata (tool name, args
+  summary) inline.
+- **HITL popup** — New modal overlay in Studio for escalate, clarify, continue, and
+  shell-permission decisions. Replaces the inline prompt pattern with a focused,
+  non-dismissible dialog that includes a visible countdown timer, action buttons,
+  and contextual metadata (affected task, agent, retry count).
+- **File inspector** — New Studio page for browsing workspace files with syntax
+  highlighting, line numbers, and a diff view against the last checkpoint snapshot.
+  Supports read-only inspection of any file in the project tree during or after
+  runs — useful for verifying worker edits without leaving the Studio.
+
 ---
 
 ## v0.10.0 — Building Blocks, Language Packs & One-Click Pipeline Switching

docs/studio.md

@@ -56,6 +56,32 @@ The loop reloads `board.json` each wave. Chaos, but *structured* chaos.
 
 ---
 
+## File inspector 🔍
+
+The **File Inspector** page lets you browse and inspect any file in the workspace
+without leaving Studio. Click any file in the tree to open it in a read-only viewer
+with syntax highlighting and line numbers.
+
+| Feature | Details |
+|---------|---------|
+| 📂 **File tree** | Full project tree — click any file to inspect |
+| 🎨 **Syntax highlighting** | Language-aware highlighting for Go, Python, JS/TS, YAML, JSON, Markdown, and more |
+| 🔢 **Line numbers** | Gutter line numbers for precise referencing |
+| 🆚 **Diff view** | Toggle to compare current file content against the last checkpoint snapshot (`rewind` data) |
+| 🔄 **Live refresh** | File content auto-refreshes when workers write changes during a run |
+| 🧊 **Read-only** | Inspection only — no accidental edits. Use the TUI or your editor to make changes |
+
+Use the File Inspector to:
+- Verify worker edits at a glance during a run
+- Spot-check generated code before the tester phase
+- Compare diffs against the pre-run checkpoint to understand what changed
+- Browse AGENTS.md, CONTEXT.md, and MEMORY.md in one place
+
+Access it from the left navigation bar — the `📄 Files` tab sits alongside Agents,
+Skills, and Blocks.
+
+---
+
 ## Human-in-the-loop modals ✋
 
 Studio blocks the relevant step with a modal (pipeline header shows **Awaiting you**):
@@ -82,6 +108,24 @@ On timeout the dedicated **@escalate** arbitrator picks retry / re-scope / abort
 
 API: `GET /api/escalate/pending` · `POST /api/escalate/answer` `{"action":"retry"}`.
 
+### HITL popup overlay
+
+v0.10.1 introduced a redesigned **HITL popup** — a modal overlay that replaces the
+old inline prompt pattern with a focused, non-dismissible dialog:
+
+| Element | Behavior |
+|---------|----------|
+| ⏱️ **Countdown timer** | Visible countdown bar showing remaining decision time; pulses red when under 10 seconds |
+| 🏷️ **Context header** | Shows affected task ID, agent name, and retry count (for escalate) |
+| 🎯 **Action buttons** | Large, color-coded buttons for each available action (approve, deny, retry, re-scope, etc.) |
+| 🚫 **Non-dismissible** | Cannot click away or close — a decision is required (or the timeout fires) |
+| 📝 **Optional note** | Text field for adding a rationale that gets logged with the decision |
+| 🔔 **Pipeline indicator** | The pipeline progress strip shows **Awaiting you** with a pulsing amber indicator |
+
+The popup appears for all HITL triggers: clarify, plan approve, escalate, continue,
+and shell permission requests. When the timeout fires, the configured fallback agent
+(e.g. `@escalate` for escalate decisions) takes over automatically.
+
 ---
 
 ## Live events (SSE) 📡

docs/testing.md

@@ -29,7 +29,10 @@ mkdir -p /tmp/slm-demo && cd /tmp/slm-demo
 printf 'package main\n\nfunc Hello() string { return "hi" }\n' > hello.go
 printf '# Agents\n\nPrefer tiny Go edits and godoc comments.\n' > AGENTS.md
 slmcode init
-slmcode blocks apply go            # apply Go pipeline + QA gate
+# Apply the Go language pack: tuned pipeline, go-worker/go-tester agents,
+# and quality gate (go vet + go test). Always run this after init so the
+# pipeline is configured for your language before the first run.
+slmcode blocks apply go
 slmcode run -v "Add a Go doc comment to Hello() explaining it returns a greeting. Keep it tiny."
 cat hello.go && slmcode board && slmcode session list
 ```