@jayjiang/byoao 1.1.2 → 2.0.0

package/dist/assets/skills/cook.md

@@ -0,0 +1,167 @@
+---
+name: cook
+description: >
+  The core knowledge compilation skill. Reads raw material (user notes, external sources)
+  and distills it into structured, cross-referenced knowledge pages in entities/, concepts/,
+  comparisons/, and queries/. Use when the user wants to compile notes into knowledge,
+  digest external material, or periodically maintain the knowledge base.
+---
+
+# /cook — Knowledge Compilation
+
+You are a knowledge compiler. Your job is to read raw material (user notes, external sources)
+and distill it into structured, cross-referenced knowledge pages.
+
+## Prerequisites Check
+
+```bash
+obsidian --version
+```
+
+If this fails, STOP and display the Obsidian CLI availability message (see /prep).
+
+## Parameters
+
+- **target** (optional): What to cook. Default: incremental (new/modified notes since last cook).
+  - `--all` or `full`: Read all user notes in the vault
+  - `"Topic Name"`: Read notes matching this keyword
+  - `path/to/note.md`: Read a specific note
+  - `<URL>`: Fetch external article and digest it
+
+## Input Scope
+
+### Incremental Mode (default)
+
+When user runs `/cook` with no arguments:
+1. Read `log.md` for last cook timestamp
+2. Scan for `.md` files outside agent directories with `modified` date after that timestamp
+3. Include any unprocessed files
+
+### Full Mode
+
+When user runs `/cook --all`:
+- Read all user notes in the vault (exclude `entities/`, `concepts/`, `comparisons/`, `queries/`)
+- Re-evaluate all entities and concepts
+
+### Targeted Mode
+
+When user runs `/cook "Feature A"` or `/cook path/to/note.md`:
+- Read only the specified notes or notes matching the keyword
+
+### External URL
+
+When user provides a URL:
+1. Fetch content using WebFetch or Obsidian Web Clipper
+2. Save as a user note in the vault (ask the user where to save, or use a sensible default like the vault root with a descriptive filename: `<slug>.md`)
+3. Add frontmatter: `title`, `source_url`, `fetched` date
+4. Process normally — the saved note becomes raw material for /cook
+
+**Note:** No dedicated `raw/` directory. External material is saved as regular user notes, consistent with the brownfield principle.
+
+## Processing Pipeline
+
+### Step 1: Read & Parse
+- Read all target notes
+- Extract frontmatter, content, wikilinks
+- Identify entities (named things), concepts (abstract ideas), decisions, contradictions
+
+### Step 2: Match Against Existing Pages
+- Check `INDEX.base` or scan `entities/`, `concepts/` for existing pages
+- Determine: create new vs. update existing
+
+### Step 3: Create/Update Pages
+- **New entities:** Create in `entities/<name>.md`
+- **New concepts:** Create in `concepts/<name>.md`
+- **Updates:** Add new information, bump `updated` date
+- **Contradictions:** Follow Update Policy
+
+**Create page thresholds:**
+- Appears in 2+ notes, OR is central subject of one note
+- Do NOT create for: passing mentions, minor details, out-of-domain topics
+
+### Step 4: Cross-Reference
+- Ensure every new/updated page has at least 2 outbound wikilinks
+- Check existing pages link back where relevant
+
+### Step 5: Update Navigation
+- `INDEX.base` auto-updates via Obsidian Base query
+- Append entry to `log.md`
+
+### Step 6: Report
+Present structured summary (see Output Report Format below).
+
+## Contradiction Handling
+
+### Detection
+- Compare claims across notes about the same entity/concept
+- Check dates — newer claims may supersede older
+- Look for explicit contradictions (e.g., "we changed from X to Y")
+
+### Resolution Workflow
+1. Note both positions with dates and source references
+2. Mark in frontmatter: `contradictions: [page-name]`
+3. Report to user with specific sources
+4. Offer to create a comparison page
+5. User decides
+
+### Update Policy
+- Newer sources generally supersede older
+- If both positions still valid (e.g., A/B testing), note both
+- Never silently overwrite — always flag for review
+
+## Output Report Format
+
+```
+Cook complete. Here's what changed:
+
+New knowledge:
+• [[feature-a]] — Response time monitoring feature
+• [[response-time-metrics]] — Why median replaced avg
+
+Updated:
+• [[zhang-san]] — Added Feature A assignment
+
+Contradiction found:
+⚠ PRD says avg(response_time) > baseline, but experiment notes say median
+  Sources: Projects/Feature-A-PRD.md vs Daily/2026-04-05.md
+  Want me to create a comparison page?
+
+Log: 1 entry added to log.md
+```
+
+**Design principles:**
+- Natural language, no technical jargon
+- Structured for quick scanning
+- Actionable (asks for decisions on contradictions)
+- Wikilinks for easy navigation
+
+## Auto-Trigger Behavior
+
+The Agent should automatically run `/cook` after:
+- Writing a note (brief report: "Cooked 1 note. Updated [[x]], created [[y]].")
+- User drops new files into the vault
+
+**When NOT to auto-trigger:**
+- Rapid-fire note creation (batch and cook once at the end)
+- `/cook` was already run in the last 5 minutes
+
+## Agent Page Identification
+
+Agent pages are identified by directory:
+| Location | Ownership |
+|----------|-----------|
+| `entities/**/*.md` | Agent |
+| `concepts/**/*.md` | Agent |
+| `comparisons/**/*.md` | Agent |
+| `queries/**/*.md` | Agent |
+| All other `.md` | User (read-only during /cook) |
+
+No `owner` frontmatter field needed.
+
+## Key Principles
+
+- **Evidence-based**: Every knowledge page cites its sources
+- **Never modify user notes**: User notes are read-only during /cook
+- **Thresholds matter**: 2+ mentions or central subject to create a page
+- **Split at 200 lines**: Break large pages into sub-topics
+- **Flag contradictions**: Never silently overwrite

package/dist/assets/skills/diagnose.md

@@ -1,72 +1,131 @@
 ---
 name: diagnose
-description: Diagnose knowledge graph health — find missing frontmatter, orphan notes, broken wikilinks, and AGENTS.md drift. Use when the user says "check my vault", "find broken links", "vault health", "what's wrong with my notes", or wants a health check on their knowledge base.
+description: >
+  Vault health check at the structural level. Checks frontmatter coverage, orphan notes,
+  broken links, AGENTS.md and SCHEMA.md drift, v2 agent directories, and overall vault configuration.
+  Broader than /health which focuses on agent pages — /diagnose checks the entire vault including user notes.
 ---
 
-# /diagnose — Knowledge Graph Health Check
+# /diagnose — Vault Diagnosis
 
-You are a vault health assistant. Your job is to diagnose issues in an Obsidian knowledge base and help the user fix them.
+You are a vault doctor. Your job is to check the overall health of the vault — structure, frontmatter coverage, configuration, and consistency across both user notes and agent pages.
 
-## Execution Flow
+## Prerequisites Check
 
-### Step 1: Locate Vault
+```bash
+obsidian --version
+```
+
+If this fails, STOP and display the Obsidian CLI availability message (see /prep).
 
-Ask the user for their vault path, or detect it from the current working directory (look for AGENTS.md in the cwd or parent directories).
+## Parameters
 
-### Step 2: Run Diagnosis
+- **focus** (optional): Specific area to check — `frontmatter`, `links`, `structure`, `config`, or `all`. Default: `all`.
 
-Call `byoao_vault_doctor` with the vault path. This runs 5 checks:
+## Process
 
-1. **Missing frontmatter** — notes without any YAML frontmatter
-2. **Missing note_type** — notes without `note_type` field (not yet woven)
-3. **Missing type/tags** — notes with frontmatter but no `type` or `tags` field
-4. **Orphan notes** — notes with no incoming or outgoing wikilinks
-5. **Broken wikilinks** — links that point to non-existent notes
+### Step 1: Frontmatter Coverage
 
-Additionally, if `INDEX.base` exists:
-6. **INDEX.base accuracy** — verify note counts match actual vault state
+```bash
+obsidian properties sort=count counts
+```
 
-### Step 3: Present Results
+Report:
+- Total notes with frontmatter vs. without
+- Most common missing fields
+- Notes with invalid frontmatter (bad dates, unknown types, etc.)
+- Tag usage: how many unique tags, how many notes per tag
 
-Format the report by severity:
+### Step 2: Broken Wikilinks
 
+Scan for wikilinks that point to non-existent files:
+
+```bash
+obsidian search "\[\[.*\]\]"
 ```
-! 3 notes without frontmatter
-  - Inbox/quick-thought.md
-  - Projects/demo-notes.md
-  - Knowledge/api-overview.md
 
-! AGENTS.md lists [[Kent]] but no People/Kent.md found
+For each wikilink found, check if the target file exists. Report broken links with:
+- Source file where the broken link appears
+- Target link that doesn't resolve
+- Suggested fix (create the missing file or remove the link)
+
+### Step 3: Orphan Detection
 
-i 2 orphan notes (no incoming or outgoing wikilinks)
-  - Archive/old-draft.md
-  - Inbox/random.md
+Find notes with no inbound wikilinks:
 
-ok 0 broken wikilinks
+```bash
+obsidian backlinks "note-name"
 ```
 
-### Step 4: Suggest Fixes
+For both user notes and agent pages, identify orphans. Note that newly created notes are expected to be orphans temporarily.
+
+### Step 4: AGENTS.md, SCHEMA.md, and v2 layout
+
+Check if `AGENTS.md` accurately reflects the current vault state:
+- Does it reference directories that no longer exist?
+- Does it miss directories that were added?
+- Are the skill references still valid?
+- Is the navigation advice still accurate?
+
+Check `SCHEMA.md`:
+- Tag taxonomy and domain sections match how tags are actually used
+- Agent directory table matches `entities/`, `concepts/`, `comparisons/`, `queries/`
+- Frontmatter expectations align with v2 `type: entity | concept | comparison | query`
+
+Verify the v2 agent directories exist and are usable: `entities/`, `concepts/`, `comparisons/`, `queries/` (note if any are missing or empty when the vault should have compiled knowledge).
+
+### Step 5: Configuration Check
 
-For each issue category, suggest a concrete next action:
+Verify vault configuration:
+- `.obsidian/` directory exists and is valid
+- `.opencode/` directory has current skill definitions
+- `SCHEMA.md` exists and has a defined tag taxonomy
+- `log.md` exists and has recent entries
+- `INDEX.base` exists for compiled knowledge discovery
 
-| Issue | Suggested Fix |
-|-------|--------------|
-| Missing frontmatter | "Run `/weave` on these files to add structure" |
-| Missing note_type | "Run `/weave` to classify and connect these notes" |
-| Missing type/tags | "Run `/weave` to fill in metadata" |
-| Orphan notes | "Consider adding `[[wikilinks]]` to connect them, or archive if unused" |
-| Broken wikilinks | "Create the target note, or fix the link name" |
-| INDEX.base stale | "Run `/wiki` to regenerate the knowledge index" |
+### Step 6: Present Diagnosis
 
-**Always ask for user confirmation before making changes.** Do not auto-fix.
+```markdown
+# Vault Diagnosis
 
-### Step 5: Update INDEX.base Timestamp
+Scanned {N} notes, {M} agent pages, {K} user notes.
 
-If `INDEX.base` exists and significant changes were made during fixes, suggest running `/wiki` to regenerate the index. If the user confirms, run `/wiki`.
+---
+
+## Frontmatter Coverage
+- Notes with frontmatter: X/Y (Z%)
+- Most common missing: {list fields}
+- Unique tags: {N} (top 5: {list})
+
+## Broken Wikilinks
+- {N} broken links found:
+  - [[target]] in [[source]] → file not found
+
+## Orphan Notes
+- {N} notes with no inbound links:
+  - [[note-name]] — consider linking from [[suggested-source]]
+
+## AGENTS.md / SCHEMA.md / layout
+- AGENTS.md: {Up to date / Needs update} — {details if outdated}
+- SCHEMA.md: {Up to date / Needs update / Missing} — {taxonomy vs usage}
+- Agent dirs (`entities/`, `concepts/`, `comparisons/`, `queries/`): {OK / Missing / Issues}
+
+## Configuration
+- .obsidian/: {OK / Missing / Issues}
+- .opencode/: {OK / Missing / Issues}
+- log.md: {OK / Missing / {N} entries, last: {date}}
+- INDEX.base: {OK / Missing / Needs update}
+
+## Overall Health
+**Score**: {Good / Fair / Needs attention}
+
+{2-3 sentence summary of the vault's overall health and the top 2-3 issues to address}
+```
 
 ## Key Principles
 
-- **Diagnose + suggest, never auto-fix**
-- **Group by severity** — warnings first, info second
-- **Actionable suggestions** — tell the user exactly what to do
-- **Respect user agency** — always ask before modifying files
+- **Comprehensive but prioritized.** Check everything, but surface the most important issues first.
+- **Actionable findings.** Every issue should come with a suggested fix.
+- **Non-destructive by default.** Report issues, don't fix them automatically.
+- **Whole vault, not just agent pages.** Unlike /health which focuses on agent-maintained directories, /diagnose checks the entire vault.
+- **Obsidian is first workbench.** All note operations go through Obsidian CLI.

package/dist/assets/skills/drift.md

@@ -1,11 +1,14 @@
 ---
 name: drift
-description: Compare stated intentions vs actual behavior over 30-60 days using daily notes and project documents. Use when the user asks "am I doing what I said I would", "what happened to my goals", "where did my time go", "check my follow-through", or wants to reflect on alignment between plans and actions.
+description: >
+  Intention-vs-action gap analysis over time. Compares what the user said they would do with
+  what actually happened. Use when the user asks "am I following through on X", "how has Y
+  changed since the plan", or wants to check if actions match intentions.
 ---
 
-# /drift — Detect Intention-Action Gaps
+# /drift — Intention vs. Action
 
-You are a behavioral analyst. Your job is to compare what the user said they would do (intentions, goals, plans) with what they actually did (daily notes, meeting notes, project updates) — revealing where actions drifted from intentions over time.
+You are an accountability mirror. Your job is to compare what the user said they would do with what actually happened — finding gaps between intentions and actions, plan vs. reality, and the slow drift of priorities over time.
 
 ## Prerequisites Check
 
@@ -13,211 +16,107 @@ You are a behavioral analyst. Your job is to compare what the user said they wou
 obsidian --version
 ```
 
-If this fails, STOP and display the Obsidian CLI availability message (see /weave for the full error text).
-
-## Tool Selection
-
-Use `obsidian` CLI for content operations (read, search, backlinks, properties, tags). Use BYOAO tools (`byoao_search_vault`, `byoao_graph_health`) when Obsidian CLI is unavailable or for graph-level structural queries.
+If this fails, STOP and display the Obsidian CLI availability message (see /prep).
 
 ## Parameters
 
-- **period** (optional): Time window to analyze. Default: "30d" (last 30 days). Accepts: "7d", "30d", "60d", "90d".
-- **focus** (optional): Limit to a project, domain, or goal. Default: all.
-- **output** (optional): Save the drift analysis as a note.
+- **topic** (optional): Specific plan, goal, or intention to track. Default: scan all recent intentions.
+- **window** (optional): Time window to analyze (e.g., "30d", "3m", "all"). Default: "30d".
 
 ## Process
 
-### Date Range Resolution
-
-Calculate the cutoff date from the `period` parameter (e.g., "30d" = today minus 30 days). Use this to filter all notes:
-- **Daily notes**: filter by filename pattern (`Daily/YYYY-MM-DD.md`)
-- **Other notes**: filter by frontmatter `date` field
-- Notes without a date: include only if they are in an active project folder or have `status: active`
+### Step 1: Find Stated Intentions
 
-### Step 1: Collect Intentions
+Search for places where the user expressed intentions:
 
-Use a structured, layered approach to find stated intentions:
-
-**Layer 1 — Structured notes (most reliable):**
 ```bash
-obsidian search "type: decision"
-obsidian search "type: plan"
+obsidian search "should" OR "need to" OR "will" OR "plan to" OR "going to" OR "decided to"
+obsidian search "goal" OR "objective" OR "target" OR "priority"
 ```
-Read sprint handoffs, project docs, and decision records within the period.
-
-**Layer 2 — Daily notes (chronological):**
-Read daily notes for the period by filename pattern (e.g., `Daily/2026-03-*.md`). Extract stated priorities, action items, and commitments.
-
-**Layer 3 — Keyword fallback (for vaults without structured frontmatter):**
-Only if Layers 1-2 produce fewer than 5 intentions:
-```bash
-obsidian search "goal"
-obsidian search "next steps"
-obsidian search "TODO"
-obsidian search "priority"
-```
-
-Also read:
-- Meeting notes with action items assigned
-- Notes with tags like `#plan`, `#goal`, `#commitment`
 
-Extract a list of **stated intentions** with dates:
-- "{date}: Planned to {X}" (source: [[Note]])
+Also check:
+- Daily notes for intention statements
+- Agent pages in `entities/` and `concepts/` for documented decisions, owners, or plans
+- Pages with `status: draft` that represent in-progress intentions
+- `log.md` as the chronological spine: cook cycles, reported changes, and stated purposes tied to dates
 
-### Step 2: Collect Actions
+### Step 2: Find Actual Actions
 
-Read daily notes and updates across the period:
+Search for evidence of what actually happened:
 
 ```bash
-obsidian search "completed"
-obsidian search "done"
-obsidian search "shipped"
-obsidian search "finished"
+obsidian search "completed" OR "done" OR "shipped" OR "implemented" OR "finished"
+obsidian search "changed" OR "switched" OR "pivoted" OR "abandoned"
 ```
 
-Read daily notes chronologically to track what actually happened:
-- What did the user write about doing?
-- What meetings happened?
-- What topics consumed attention?
+Check:
+- Recent daily notes for actual activities
+- Agent pages in `entities/` and `concepts/` for current state and decision descriptions
+- `log.md` entries since the intention was stated
+- Updated frontmatter dates and `status` changes
+- New pages created vs. pages left in draft
 
-Extract a list of **actual actions** with dates:
-- "{date}: Did {Y}" (source: [[Daily Note]])
+### Step 3: Compare Intentions to Actions
 
-### Step 3: Alignment Analysis
+For each intention found:
 
-For each stated intention, check:
-
-| Status | Meaning |
-|--------|---------|
-| **Aligned** | Intention was followed through with documented evidence |
-| **Delayed** | Work started but timeline slipped |
-| **Drifted** | Work went in a different direction than planned |
-| **Abandoned** | Intention was stated but never acted on |
-| **Emergent** | Action happened that was never planned (reactive work) |
+1. **Followed through** — Evidence shows the action was taken as planned
+2. **Partially followed** — Some action was taken but not fully
+3. **Deferred** — Still planned but not yet acted on
+4. **Diverged** — Action was taken but in a different direction
+5. **Abandoned** — No evidence of any action
 
 ### Step 4: Identify Drift Patterns
 
-Look for systemic patterns, not just individual misses:
-
-**Priority displacement** — Planned work was consistently displaced by reactive work.
-> "In 4 of 5 weeks, the Friday daily note mentions not getting to the planned work because of urgent requests."
-
-**Scope creep** — The scope of a project expanded without acknowledgment.
-> "The original plan in [[Project Plan]] had 5 deliverables. Current notes reference 9, but no re-planning happened."
-
-**Energy leaks** — Time going to undocumented work.
-> "Daily notes from weeks 3-5 rarely mention the stated priority. The gap suggests time is going somewhere not reflected in the vault."
-**Caveat:** This pattern is only meaningful if the user writes daily notes consistently. If daily notes are sparse or irregular, the absence of mentions is a documentation gap, not evidence of time spent elsewhere. Note this distinction in the report.
-
-**Goal abandonment** — Goals that silently disappeared.
-> "The Q1 goal of 'improve test coverage' was mentioned 3 times in January and never again."
+Look for systematic patterns:
 
-**Emergent priorities** — Unplanned work that became dominant.
-> "'Customer escalations' wasn't in any plan but appears in 60% of daily notes."
+- **Priority drift**: The user said X was top priority, but most time went to Y
+- **Scope drift**: A small intention grew into a much larger effort (or shrank)
+- **Direction drift**: The approach changed from the original plan
+- **Timeline drift**: Things took significantly longer (or shorter) than expected
+- **Attention drift**: An intense focus faded and wasn't replaced by anything
 
-### Step 5: Present the Analysis
+### Step 5: Present the Drift Report
 
 ```markdown
-# Drift Analysis: {period}
+# Drift Report: {topic or "Recent Intentions"}
 
-Comparing intentions vs actions over {period}, focused on {focus or "all areas"}.
-
-## Summary
-
-| Category | Count |
-|----------|-------|
-| Aligned | {N} intentions followed through |
-| Delayed | {N} intentions behind schedule |
-| Drifted | {N} intentions changed direction |
-| Abandoned | {N} intentions never acted on |
-| Emergent | {N} unplanned actions taken |
-
-**Overall drift score**: {Low / Moderate / High}
+Analyzed {N} notes from {start date} to {end date}.
 
 ---
 
-## Aligned (on track)
-
-- **{Intention}** ({date})
-  Stated in: [[Plan Note]]
-  Evidence: [[Action Note 1]], [[Action Note 2]]
+## Followed Through ✅
+- **{intention}** — {what was done, evidence from [[note]]}
 
-## Delayed
+## Partially Followed ⚡
+- **{intention}** — {what was done vs. what was planned, gap evidence from [[note]]}
 
-- **{Intention}** ({original date} → still in progress)
-  Stated in: [[Plan Note]]
-  Last mention: [[Recent Note]] ({date})
-  **Likely cause**: {what the notes suggest}
+## Deferred ⏳
+- **{intention}** — {stated on [[date]] in [[note]], no action found since}
 
-## Drifted
+## Diverged ↩
+- **{intention}** — {original plan from [[note A]], actual outcome from [[note B]]}
 
-- **{Intention}** → became **{what it turned into}**
-  Original: [[Plan Note]] — "{original plan}"
-  Current: [[Recent Note]] — "{what's actually happening}"
-  **The shift**: {when and why direction changed}
-
-## Abandoned
-
-- **{Intention}** ({date stated}, last mentioned {date})
-  Stated in: [[Note]]
-  **No evidence of**: follow-up, cancellation decision, or handoff
-  **Question**: Was this a conscious decision or did it just fade?
-
-## Emergent (unplanned)
-
-- **{Action pattern}** — appeared in {N} daily notes, not in any plan
-  **Impact**: {how much time/attention this consumed}
-  **Question**: Should this be formally planned/resourced?
+## Abandoned ❌
+- **{intention}** — {stated on [[date]], zero evidence of action}
 
 ---
 
 ## Drift Patterns
 
-### {Pattern name}
-{Description with evidence from notes}
-
----
-
-## Reflections
-
-These questions are for your consideration — not judgments:
-
-1. {Question about the most significant drift}
-2. {Question about whether emergent work should be planned}
-3. {Question about abandoned intentions}
+### Priority Drift
+{Evidence that stated priorities don't match actual time allocation}
 
-## Suggested Actions
+### Direction Drift
+{Evidence that the approach changed from the original plan}
 
-- Write a decision record for {abandoned intention} — cancel formally or recommit
-- Add {emergent pattern} to the next planning cycle
-- Run `/trace topic="{drifted topic}"` to understand the shift
-- Re-read [[original plan]] with fresh eyes
+## Overall Assessment
+{2-3 sentences: Is the user generally following through on intentions? Where is the biggest gap? Is the drift a problem or a healthy adaptation?}
 ```
 
-### Step 6: Save (Optional)
-
-At the end of your analysis, ask:
-
-> "Would you like me to save this as a note?"
-
-If the user confirms, save with frontmatter:
-
-```yaml
----
-title: "Drift Analysis: {period}"
-note_type: literature
-type: analysis
-date: <today>
-tags: [drift, reflection]
----
-```
-
-Use `obsidian create` to save. Ask the user where they'd like it saved.
-
 ## Key Principles
 
-- **Descriptive, not judgmental**: "You said X and did Y" is fine. "You failed to do X" is not. Drift is information, not failure.
-- **Vault evidence only**: Only compare what's documented. If something isn't in the vault, it's a gap in documentation, not proof of inaction.
-- **Surface patterns, not incidents**: One missed item is noise. Three missed items in the same category is a pattern worth examining.
-- **Respect the user's agency**: Drift may be intentional adaptation, not failure. Present findings neutrally and let the user interpret.
-- **Emergent work is valid**: Unplanned work that was important should be recognized, not treated as distraction.
+- **Factual, not judgmental.** Report the gap between intention and action without moralizing. The user decides if it matters.
+- **Evidence-based.** Every drift claim must cite specific notes showing both the intention and the actual outcome.
+- **Drift isn't always bad.** Sometimes changing direction is the right call. Flag the drift; let the user judge.
+- **Obsidian is first workbench.** All note operations go through Obsidian CLI.