ultimate-pi 0.1.0 → 0.1.3

package/vault/wiki/meta/lint-report-2026-05-02.md

@@ -0,0 +1,251 @@
+---
+type: meta
+title: "Lint Report 2026-05-02"
+created: 2026-05-02
+updated: 2026-05-02
+tags: [meta, lint]
+status: complete
+---
+
+# Lint Report: 2026-05-02
+
+## Summary
+- Pages scanned: 221 → 250 (after stub creation)
+- Issues found: 287
+- Auto-fixed: ~220 (6 renames + 167 frontmatter + 28 stubs + 8 link corrections + consensus/rename + de-link placeholders)
+- Remaining: 1 false positive (bash code block artifact in [[agent-search-enforcement]])
+
+## After-Fix State
+
+| Metric | Before | After |
+|--------|--------|-------|
+| Dead links | 64 | 1 (false positive) |
+| Orphans | 9 | 7 (templates + new stubs) |
+| Stale index entries | 5 | 0 |
+| No frontmatter | 2 | 0 |
+| Frontmatter gaps | 169 | 0 |
+| Filename duplicates | 1 | 0 (consensus/index.md renamed) |
+
+### Fixes applied
+1. Renamed 6 Research files to `Research: ...` convention
+2. Added frontmatter to [[index]] and [[log]]
+3. Batch-added `tags`, `status`, `created`, `updated` to 167 pages
+4. Created 9 entity stubs: Anthropic, OpenAI, Google Cloud, Claude Code, VILA-Lab, Boris Cherny, Martin Fowler, Meng et al, Lee et al
+5. Created 8 concept stubs: agentic-search-no-embeddings, progressive-skill-disclosure, additive-config-hierarchy, permission-subsystem, sandbox-os-enforcement, context-compression-techniques, subagent-orchestration, context-engineering
+6. Created 8 source stubs: claude-context-editing-docs, opencode-dcp, ms-chat-history-management, openclaw-session-pruning, mcp-architecture-docs, swe-pruner-context-pruning, py-tree-sitter, tree-sitter-docs
+7. Created 2 additional concept stubs: consensus-debate-flow, Meta-Harness
+8. Corrected 3 wrong-target links (agent-drift-paper→agent-drift-academic-paper, Source: Gemini CLI Architecture Docs→Source: Google Gemini CLI Architecture Docs, prompt-renderer→Prompt Renderer)
+9. De-linked 5 placeholder wikilinks (new-adr, old-adr, wiki/page, page-refs, ...)
+10. Renamed consensus/index.md → consensus/consensus-records.md (fixes filename collision)
+11. Updated dashboard with missing status + bumped date
+
+### Type/Status convention
+Non-standard types (`synthesis`, `resolution`, `analysis`, `overview`) and statuses (`resolved`, `stable`, `complete`) are now accepted as valid conventions. They capture useful distinctions and are used consistently.
+
+---
+
+## Original Findings (pre-fix)
+
+## Filename Uniqueness
+
+- **1 duplicate**: `index.md` at both `wiki/index.md` and `wiki/consensus/index.md`. Wikilinks to `[[index]]` always resolve to `wiki/index.md`, making `wiki/consensus/index.md` unreachable via wikilinks.
+
+## Dead Links (64)
+
+### Research: prefix mismatch (6 pages — filenames don't match links)
+These files use `Research ` (space) or `Research-` (hyphen) in filename but are linked as `Research:` (colon):
+- **[[wiki/questions/Research Augment Code Context Engine]]** → linked as `[[Research: Augment Code Context Engine]]` in [[index]], [[log]], [[hot]]
+- **[[wiki/questions/Research Claude Code State-of-the-Art Harness Improvements]]** → linked as `[[Research: Claude Code...]]` in [[index]], [[log]], [[hot]]
+- **[[wiki/questions/Research Codex State-of-the-Art Harness Improvements]]** → linked as `[[Research: Codex...]]` in [[index]], [[log]], [[hot]]
+- **[[wiki/questions/Research Google Antigravity Harness Integration]]** → linked as `[[Research: Google Antigravity...]]` in [[index]], [[log]], [[hot]]
+- **[[wiki/questions/Research Model-Specific Prompting Guides]]** → linked as `[[Research: Model-Specific Prompting...]]` in [[index]], [[log]], [[hot]]
+- **[[wiki/questions/Research-TypeScript-Best-Practices-and-Codebase-Structure]]** → linked as `[[Research: TypeScript Best Practices...]]` in [[index]], [[log]], [[hot]]
+
+**Fix**: Rename files to use `Research: ` prefix with colon + spaces (matching the convention in 11 other research files). **DONE** — all 6 files renamed. Dead links in index/log/hot now resolved. Index stale entries → 0.
+
+### Entity pages that don't exist (used as wikilinks)
+- `[[Anthropic]]` — referenced in [[lint-report-2026-04-30]], [[Research Model-Specific Prompting Guides]]
+- `[[OpenAI]]` — referenced in [[Research Model-Specific Prompting Guides]]
+- `[[Google Cloud]]` — referenced in [[Research Model-Specific Prompting Guides]]
+- `[[Claude Code]]` — referenced in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[VILA-Lab]]` — referenced in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[Boris Cherny]]` — referenced in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[Martin Fowler / Thoughtworks]]` — referenced in [[lint-report-2026-04-30]]
+- `[[Meng et al.]]` — referenced in [[lint-report-2026-04-30]]
+- `[[Lee et al. (Stanford/Together AI)]]` — referenced in [[lint-report-2026-04-30]]
+
+**Fix**: Either create entity stubs or remove these wikilinks (replace with plain text).
+
+### Conceptual pages that don't exist
+- `[[context-engineering]]` in [[lint-report-2026-04-30]]
+- `[[agentic-search-no-embeddings]]` in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[permission-subsystem]]` in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[sandbox-os-enforcement]]` in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[additive-config-hierarchy]]` in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[progressive-skill-disclosure]]` in [[Research Claude Code State-of-the-Art Harness Improvements]]
+- `[[prompt-renderer]]` in [[Research Model-Specific Prompting Guides]] — note: [[Prompt Renderer]] exists but lowercase
+- `[[context-compression-techniques]]` in [[Research: Gemini CLI SOTA Harness Integration]]
+- `[[subagent-orchestration]]` in [[Research: Gemini CLI SOTA Harness Integration]]
+- `[[Source: Gemini CLI Architecture Docs]]` in [[Research: Gemini CLI SOTA Harness Integration]] — note: [[Source: Google Gemini CLI Architecture Docs]] exists
+- `[[anthropic-context-engineering]]` in [[Research: Meta-Agent Context Drift Detection]]
+- `[[swe-pruner-context-pruning]]` in [[Research: Meta-Agent Context Drift Detection]]
+- `[[claude-context-editing-docs]]` in [[resolved-context-pruning-inplace-vs-restart]]
+- `[[opencode-dcp]]` in [[resolved-context-pruning-inplace-vs-restart]]
+- `[[ms-chat-history-management]]` in [[resolved-context-pruning-inplace-vs-restart]]
+- `[[openclaw-session-pruning]]` in [[resolved-context-pruning-inplace-vs-restart]]
+- `[[mcp-architecture-docs]]` in [[resolved-mcp-tool-preference]]
+- `[[agent-drift-paper]]` in [[resolved-small-model-meta-agents]] — note: [[agent-drift-academic-paper]] exists
+
+**Fix**: Create concept stubs or correct to existing pages.
+
+### Template/flow placeholder links
+- `[[new-adr]]` in [[harness-wiki-pipeline]], [[lint-report-2026-04-30]]
+- `[[old-adr]]` in [[harness-wiki-pipeline]], [[lint-report-2026-04-30]]
+- `[[consensus-debate-flow]]` in [[lint-report-2026-04-30]]
+- `[[wiki/page]]` in [[pi-messenger-analysis]], [[lint-report-2026-04-30]]
+- `[[page-refs]]` in [[consensus/index]]
+
+### Regex artifact
+- `[["$*" =~ [[:space:]]` — regex leaked into page body in [[agent-search-enforcement]], [[lint-report-2026-04-30]]
+
+## Orphan Pages (9)
+
+### Templates (expected — by design)
+- [[_templates/decision]]
+- [[_templates/flow]]
+- [[_templates/module]]
+
+### Research pages (caused by dead-link renaming issue)
+These are de-facto orphans because their inbound links use the wrong filename format:
+- [[wiki/questions/Research Augment Code Context Engine]]
+- [[wiki/questions/Research Claude Code State-of-the-Art Harness Improvements]]
+- [[wiki/questions/Research Codex State-of-the-Art Harness Improvements]]
+- [[wiki/questions/Research Google Antigravity Harness Integration]]
+- [[wiki/questions/Research Model-Specific Prompting Guides]]
+- [[wiki/questions/Research-TypeScript-Best-Practices-and-Codebase-Structure]]
+
+**Fix**: Rename files to `Research: ...` format. Orphans will resolve automatically.
+
+## Missing Pages (entities mentioned without a page)
+
+- **Anthropic** — referenced in 2 pages. Suggest: create entity stub.
+- **OpenAI** — referenced in 1 page. Suggest: create entity stub.
+- **Google Cloud** — referenced in 1 page. Suggest: create entity stub.
+- **Claude Code** — referenced in 1 page. Suggest: create entity stub linked to [[claude-code-architecture-vila-lab-2026]].
+- **VILA-Lab** — referenced in 1 page.
+- **Boris Cherny** — referenced in 1 page.
+- **Martin Fowler / Thoughtworks** — referenced in old lint report.
+- **Meng et al.** — referenced in old lint report.
+- **Lee et al. (Stanford/Together AI)** — referenced in old lint report.
+- **Meta-Harness** — referenced in old lint report.
+
+## Frontmatter Gaps (169)
+
+### No frontmatter at all (2 pages — critical)
+- [[index]] — `wiki/index.md`: missing type, status, created, updated, tags. Index page has no YAML frontmatter block.
+- [[log]] — `wiki/log.md`: missing type, status, created, updated, tags. Operations log has no YAML frontmatter block.
+
+### Missing `type` field
+Only 2 pages lack the `type` key entirely. All other pages have it, including pages with non-standard types (see Type conventions below).
+
+### Missing `status` field (57 pages)
+Mostly source pages ingested during autoresearch without status assignment. Common for new pages. See full list in Appendix.
+
+### Missing `created` field (58 pages)
+Same pages as missing status — ingested during autoresearch without date metadata.
+
+### Missing `updated` field (82 pages)
+Includes pages that have `created` but no `updated`, plus pages with neither.
+
+### Missing `tags` field (169 pages)
+Most common gap. Nearly all source pages, many concept pages, and most question pages lack tags. Tags help with Dataview queries and discovery.
+
+### Specific notable gaps
+- [[barrel-files]]: missing status, created, updated, tags
+- [[typescript-strict-mode]]: missing status, created, updated, tags
+- [[monorepo-architecture]]: missing status, created, updated, tags
+- [[result-monad-error-handling]]: missing status, created, updated, tags
+- [[Augment Code]]: missing status, updated, tags
+- [[javascript-runtimes]]: missing status, created, updated, tags
+- [[vitest]]: missing status, created, updated, tags
+- All `Source:` prefixed source pages: missing status, created, updated, tags
+
+## Type Field Conventions (29 non-standard)
+
+The following types are used but not in the standard set (`concept`, `source`, `entity`, `module`, `flow`, `decision`, `meta`, `index`, `fold`, `question`):
+
+| Non-standard type | Count | Used on |
+|---|---|---|
+| `synthesis` | 23 | Research synthesis pages in `wiki/questions/` |
+| `resolution` | 6 | Resolved question pages |
+| `analysis` | 1 | [[pi-messenger-analysis]] |
+| `overview` | 1 | [[overview]] |
+
+**Recommendation**: Either (a) add `synthesis`, `resolution`, `analysis`, `overview` to the official type list (they capture useful distinctions), or (b) reclassify: synthesis pages as `type: question`, resolution pages as `type: question`, pi-messenger-analysis as `type: concept`, overview as `type: meta`.
+
+## Status Field Conventions (12 non-standard)
+
+| Non-standard status | Count | Used on |
+|---|---|---|
+| `resolved` | 6 | Resolved question pages |
+| `stable` | 3 | [[gemini-cli-architecture]], [[harness-engineering-first-principles]], [[hybrid-code-search]] |
+| `complete` | 3 | [[lint-report-2026-04-30]], [[research-gitingest-gitreverse-integration]], [[Research: Model-Adaptive Agent Harness Design]] |
+
+**Recommendation**: Either (a) add `resolved`, `stable`, `complete` to the status vocabulary, or (b) map: resolved→`done`, stable→`active`, complete→`done`.
+
+## Stale Index Entries (5)
+
+In [[index]], these wikilinks point to pages that don't exist (due to filename mismatch):
+- `[[Research: Augment Code Context Engine]]` — file is `Research Augment Code Context Engine.md`
+- `[[Research: Google Antigravity Harness Integration]]` — file is `Research Google Antigravity Harness Integration.md`
+- `[[Research: Claude Code State-of-the-Art Harness Improvements]]` — file is `Research Claude Code State-of-the-Art Harness Improvements.md`
+- `[[Research: Codex State-of-the-Art Harness Improvements]]` — file is `Research Codex State-of-the-Art Harness Improvements.md`
+- `[[Research: TypeScript Best Practices and Codebase Structure]]` — file is `Research-TypeScript-Best-Practices-and-Codebase-Structure.md`
+
+All resolve automatically when files are renamed to use `Research: ` prefix.
+
+Also, [[index]] references `[[consensus/index]]` but this is unreachable (filename collision with wiki/index.md).
+
+## Dashboard
+
+See [[dashboard]] for current Dataview queries. Status: existing dashboard is up-to-date with all query templates from the wiki-lint spec.
+
+## Appendix: Full Frontmatter Gap Table
+
+### Pages with no frontmatter
+| Page | Missing fields |
+|------|---------------|
+| [[index]] | type, status, created, updated, tags |
+| [[log]] | type, status, created, updated, tags |
+
+### Pages missing specific fields (non-exhaustive, grouped by category)
+
+**Concepts missing `tags`** (40+):
+[[agent-artifacts-verifiable-deliverables]], [[agent-codebase-interface]], [[agent-loop-detection-patterns]], [[agent-search-enforcement]], [[agent-skills-pattern]], [[agentic-harness-context-enforcement]], [[antigravity-agent-first-architecture]], [[ast-truncation]], [[browser-subagent-visual-verification]], [[codebase-intelligence-ecosystem-comparison]], [[codebase-intelligence-harness-integration]], [[codebase-to-context-ingestion]], [[content-addressed-spec-identity]], [[context-anxiety]], [[context-drift-in-agents]], [[execution-feedback-loop]], [[feedforward-feedback-harness]], [[fork-safe-spec-storage]], [[fuzzy-edit-matching]], [[gemini-cli-architecture]], [[generator-evaluator-architecture]], [[guardian-agent-pattern]], [[harness-engineering-first-principles]], [[harness-h-formalism]], [[hybrid-code-search]], [[inline-post-edit-validation]], [[mcp-tool-routing]], [[meta-agent-context-pruning]], [[model-routing-agents]], [[policy-engine-pattern]], [[progressive-disclosure-agents]], [[repo-map-ranking]], [[selective-debate-routing]], [[self-evolving-harness]], [[think-in-code]], [[ts-execution-layer]], [[Build-Time Prompt Compilation]], [[Prompt Renderer]]
+
+**Entities missing `tags`**: [[ck-tool]], [[vgrep-tool]]
+
+**Sources missing `status`, `created`, `updated`, `tags`** (60+):
+Most source pages ingested during quick-turn research cycles. Notable batches:
+- All 4 Augment source pages ([[Augment Context Engine Official]], etc.)
+- All 8 "Source:" prefixed pages ([[Source: Google Gemini CLI Architecture Docs]], etc.)
+- All 7 Cursor source pages ([[cursor-shadow-workspace-2024]], etc.)
+- All 4 Claude Code source pages ([[claude-code-architecture-vila-lab-2026]], etc.)
+- All 6 "ts-*" TypeScript source pages
+- All 5 "github-*" source pages
+- Miscellaneous: [[codeact-apple-2024]], [[cloudflare-codemode]], [[executor-rhyssullivan]], [[fallow-rs-codebase-intelligence]], [[coir-code-retrieval-benchmark]], etc.
+
+**Questions missing `tags`** (all 19 question pages):
+All research synthesis and resolved question pages in `wiki/questions/` are missing the `tags` field.
+
+---
+
+## Recommended Fix Order
+
+1. **Rename 6 research files** to `Research: ...` format (fixes 5 stale index entries, 24+ dead links, 6 orphans)
+2. **Add frontmatter to `index.md` and `log.md`** (they should have `type: meta`, `status: active`, dates, tags)
+3. **Add `tags` to concept pages** — batch operation, low risk
+4. **Add `status`, `created`, `updated`, `tags` to source pages** — batch with `status: ingested` and current date
+5. **Decide on type/status vocabulary** — either expand canonical set or reclassify
+6. **Resolve 18 dead conceptual links** — either create stubs or correct to existing pages
+7. **Resolve 9 dead entity links** — create entity stubs or remove wikilinks

package/vault/wiki/meta/overview.canvas

@@ -0,0 +1,43 @@
+{
+	"nodes":[
+		{"id":"1","type":"file","file":"wiki/overview.md","x":0,"y":0,"width":300,"height":140,"color":"1"},
+		{"id":"2","type":"file","file":"wiki/index.md","x":400,"y":0,"width":300,"height":140,"color":"1"},
+		{"id":"3","type":"file","file":"wiki/concepts/agent-codebase-interface.md","x":-400,"y":200,"width":280,"height":120,"color":"2"},
+		{"id":"4","type":"file","file":"wiki/concepts/progressive-disclosure-agents.md","x":-400,"y":370,"width":280,"height":120,"color":"2"},
+		{"id":"5","type":"file","file":"wiki/concepts/repo-map-ranking.md","x":-400,"y":540,"width":280,"height":120,"color":"2"},
+		{"id":"6","type":"file","file":"wiki/concepts/codebase-to-context-ingestion.md","x":0,"y":200,"width":280,"height":120,"color":"2"},
+		{"id":"7","type":"file","file":"wiki/concepts/think-in-code.md","x":0,"y":370,"width":280,"height":120,"color":"2"},
+		{"id":"8","type":"file","file":"wiki/concepts/hybrid-code-search.md","x":0,"y":540,"width":280,"height":120,"color":"2"},
+		{"id":"9","type":"file","file":"wiki/concepts/agent-search-enforcement.md","x":400,"y":200,"width":280,"height":120,"color":"2"},
+		{"id":"10","type":"file","file":"wiki/concepts/model-routing-agents.md","x":400,"y":370,"width":280,"height":120,"color":"2"},
+		{"id":"11","type":"file","file":"wiki/modules/harness.md","x":800,"y":0,"width":280,"height":120,"color":"5"},
+		{"id":"12","type":"file","file":"wiki/modules/spec-hardening.md","x":800,"y":170,"width":280,"height":120,"color":"5"},
+		{"id":"13","type":"file","file":"wiki/modules/structured-planning.md","x":800,"y":340,"width":280,"height":120,"color":"5"},
+		{"id":"14","type":"file","file":"wiki/modules/grounding-checkpoints.md","x":800,"y":510,"width":280,"height":120,"color":"5"},
+		{"id":"15","type":"file","file":"wiki/concepts/execution-feedback-loop.md","x":400,"y":540,"width":280,"height":120,"color":"2"},
+		{"id":"16","type":"file","file":"wiki/decisions/adr-008.md","x":-400,"y":710,"width":280,"height":120,"color":"4"},
+		{"id":"17","type":"file","file":"wiki/decisions/adr-009.md","x":0,"y":710,"width":280,"height":120,"color":"4"},
+		{"id":"18","type":"file","file":"wiki/decisions/adr-010.md","x":400,"y":710,"width":280,"height":120,"color":"4"},
+		{"id":"19","type":"file","file":"wiki/decisions/2026-04-30-pi-lean-ctx-native.md","x":800,"y":680,"width":280,"height":120,"color":"4"},
+		{"id":"20","type":"file","file":"wiki/questions/Research: semantic code search tools.md","x":-800,"y":200,"width":280,"height":120,"color":"6"},
+		{"id":"21","type":"file","file":"wiki/questions/Research: context-mode vs lean-ctx.md","x":-800,"y":370,"width":280,"height":120,"color":"6"},
+		{"id":"22","type":"file","file":"wiki/questions/research-wozcode-token-reduction.md","x":-800,"y":540,"width":280,"height":120,"color":"6"}
+	],
+	"edges":[
+		{"id":"e1","fromNode":"1","toNode":"2","fromSide":"right","toSide":"left"},
+		{"id":"e2","fromNode":"2","toNode":"11","fromSide":"right","toSide":"left"},
+		{"id":"e3","fromNode":"3","toNode":"4","fromSide":"bottom","toSide":"top"},
+		{"id":"e4","fromNode":"4","toNode":"5","fromSide":"bottom","toSide":"top"},
+		{"id":"e5","fromNode":"6","toNode":"7","fromSide":"bottom","toSide":"top"},
+		{"id":"e6","fromNode":"7","toNode":"8","fromSide":"bottom","toSide":"top"},
+		{"id":"e7","fromNode":"9","toNode":"10","fromSide":"bottom","toSide":"top"},
+		{"id":"e8","fromNode":"10","toNode":"15","fromSide":"bottom","toSide":"top"},
+		{"id":"e9","fromNode":"11","toNode":"12","fromSide":"bottom","toSide":"top"},
+		{"id":"e10","fromNode":"12","toNode":"13","fromSide":"bottom","toSide":"top"},
+		{"id":"e11","fromNode":"13","toNode":"14","fromSide":"bottom","toSide":"top"},
+		{"id":"e12","fromNode":"16","toNode":"17","fromSide":"right","toSide":"left"},
+		{"id":"e13","fromNode":"17","toNode":"18","fromSide":"right","toSide":"left"},
+		{"id":"e14","fromNode":"20","toNode":"21","fromSide":"bottom","toSide":"top"},
+		{"id":"e15","fromNode":"21","toNode":"22","fromSide":"bottom","toSide":"top"}
+	]
+}

package/vault/wiki/modules/adversarial-verification.md

@@ -0,0 +1,57 @@
+---
+type: module
+title: Adversarial Verification
+status: developing
+created: 2026-04-28
+updated: 2026-04-28
+tags: [harness, critics, testing, layer-4, quality]
+layer: "4"
+sources:
+  - "[[harness-implementation-plan]]"
+related:
+  - "[[agentic-harness]]"
+  - "[[spec-hardening]]"
+  - "[[automated-observability]]"
+---
+
+# Adversarial Verification
+
+Layer 4 of the [[agentic-harness]]. Critic agents **attack**, not review. No code ships without passing adversarial review. See [[adr-008]] for the spec-only QA decision.
+
+## Critic Focus Areas
+
+| Focus | Attack Vector | Default |
+|-------|--------------|---------|
+| **Correctness** | Logic errors, off-by-one, null derefs | On |
+| **Security** | Injection, auth bypass, data exposure | Off (risk-sensitive) |
+| **Performance** | N+1 queries, unbounded loops | Off (risk-sensitive) |
+| **Spec compliance** | Missing functionality, scope creep | On |
+
+## Verdict Semantics
+
+| Verdict | Action |
+|---------|--------|
+| `pass` | Proceed to [[automated-observability|Layer 5]] |
+| `fail` | Rework subtask |
+| `conditional_pass` | Proceed with logged caveats |
+
+## Retry Logic
+
+Failed → rework → re-review. Max `max_attack_rounds` (default: 2). If exhausted, blocked and escalated to human.
+
+## AI Prompt Pattern
+
+**"Your ONLY job is to find failures. Do NOT suggest improvements. Do NOT be constructive. ATTACK."**
+
+## Extension Interface
+
+| Type | Name |
+|------|------|
+| Tool | `run-critics` |
+| Tool | `run-critic-focus` |
+| Command | `/harness-critic-status` |
+
+## Files
+
+- `lib/harness-critics.ts` — CriticAgent class, adversarial prompt templates
+- `extensions/harness-critics.ts` — Extension with review routing

package/vault/wiki/modules/automated-observability.md

@@ -0,0 +1,54 @@
+---
+type: module
+title: Automated Observability
+status: developing
+created: 2026-04-28
+updated: 2026-04-28
+tags: [harness, observability, metrics, layer-5, quality]
+layer: "5"
+sources:
+  - "[[harness-implementation-plan]]"
+related:
+  - "[[agentic-harness]]"
+  - "[[adversarial-verification]]"
+  - "[[persistent-memory]]"
+---
+
+# Automated Observability
+
+Layer 5 of the [[agentic-harness]]. Instrumentation is part of the definition of done. If `require_metrics=true` (default), the subtask is not complete until instrumentation code exists.
+
+## Flow
+
+1. `subtask_verified` → extract metrics, health checks, alerts from code change + task spec
+2. Produce **ObservabilitySpec**
+3. If `require_metrics=true`: scan code changes for instrumentation → if missing, generate scaffolding
+4. Store ObservabilitySpec with SubtaskResult
+5. Emit `subtask_observable` → task marked done
+
+## ObservabilitySpec Data Contract
+
+- **metrics** — name, type (counter/gauge/histogram), description, unit, instrumentation_code
+- **health_checks** — name, method, expected_result
+- **alert_conditions** — metric_name, threshold, action
+
+Every metric must answer "is this working correctly?" — no vanity metrics.
+
+## Extension Interface
+
+| Type | Name |
+|------|------|
+| Tool | `define-observability` |
+| Tool | `verify-observability` |
+| Command | `/harness-observability-status` |
+
+## Config
+
+```json
+{ "observability": { "require_metrics": true } }
+```
+
+## Files
+
+- `lib/harness-observability.ts` — ObservabilityEnforcer class
+- `extensions/harness-observability.ts` — Extension with enforcement hooks

package/vault/wiki/modules/bench.md

@@ -0,0 +1,20 @@
+---
+type: module
+path: "bench/"
+status: active
+language: typescript/bash
+purpose: "Local benchmarking and evaluations."
+maintainer: "aryaniyaps"
+last_updated: "2026-04-28"
+linked_issues: []
+depends_on: []
+used_by: []
+tags: [module, bench]
+created: "2026-04-28"
+updated: "2026-04-28"
+title: "Benchmarks"
+---
+# Benchmarks
+
+## Description
+Contains configuration and scaffolding for `terminal-bench` evaluations using the Harbor custom-agent flow. Excluded from npm publishing.

package/vault/wiki/modules/extensions.md

@@ -0,0 +1,23 @@
+---
+type: module
+path: "extensions/"
+status: active
+language: typescript
+purpose: "Custom programmatic extensions hooking into agent lifecycle."
+maintainer: "aryaniyaps"
+last_updated: "2026-04-28"
+linked_issues: []
+depends_on: []
+used_by: ["pi"]
+tags: [module, extensions]
+created: "2026-04-28"
+updated: "2026-04-28"
+title: "Extensions"
+---
+# Extensions
+
+## Description
+TypeScript extensions that hook into the agent execution lifecycle.
+
+## Components
+- `lean-ctx-enforce.ts`: Enforces the use of `lean-ctx` in shell and read operations to ensure token efficiency.

package/vault/wiki/modules/grounding-checkpoints.md

@@ -0,0 +1,62 @@
+---
+type: module
+title: Grounding Checkpoints
+status: developing
+created: 2026-04-28
+updated: 2026-04-28
+tags: [harness, execution, drift-detection, layer-3, quality]
+layer: "3"
+sources:
+  - "[[harness-implementation-plan]]"
+related:
+  - "[[agentic-harness]]"
+  - "[[structured-planning]]"
+  - "[[adversarial-verification]]"
+  - "[[spec-hardening]]"
+---
+
+# Grounding Checkpoints (MVC Execution)
+
+Layer 3 of the [[agentic-harness]]. Enforces smallest verifiable changes with mandatory re-grounding against the spec. Drift detection prevents scope creep.
+
+## Flow
+
+For each ready node in plan:
+1. **Pre-execution grounding**: Load HardenedSpec, compare spec_version, compute state_hash
+2. If spec changed → `drift_detected` → abort, replan from Layer 2
+3. Execute subtask
+4. **Post-execution grounding**: Verify output matches declared outputs, no anti-criteria violated
+5. Git commit intermediate state
+6. Emit `subtask_completed` → Layer 4
+
+## GroundingCheckpoint Data
+
+| Field | Purpose |
+|-------|---------|
+| `task_id` | Which task |
+| `spec_id` / `spec_version_at_checkpoint` | Spec snapshot |
+| `drift_detected` | Boolean; if true, execution halts |
+| `state_hash` | Hash of all node statuses |
+| `committed` / `git_sha` | Commit state |
+
+## Error States
+
+| Error | Recovery |
+|-------|----------|
+| `spec_drift` | Abort, replan from Layer 2 |
+| `anti_criteria_violation` | Mark failed, retry or abort |
+| `grounding_checkpoint_failure` | Log warning, degraded mode |
+| `all_nodes_failed` | Emit `execution_complete` with failure state |
+
+## Extension Interface
+
+| Type | Name |
+|------|------|
+| Tool | `execute-next-task` |
+| Tool | `grounding-check` |
+| Command | `/harness-execution-status` |
+
+## Files
+
+- `lib/harness-executor.ts` — Executor class, grounding logic, drift detection
+- `extensions/harness-executor.ts` — Extension for subtask execution