flonat-research 0.1.0

package/skills/system-audit/SKILL.md

@@ -0,0 +1,157 @@
+---
+name: system-audit
+description: "Use when you need to run parallel audits across skills, hooks, agents, rules, and conventions."
+allowed-tools: Bash(ls*), Bash(readlink*), Bash(wc*), Bash(git*), Bash(test*), Bash(stat*), Bash(find*), Read, Glob, Grep, Task
+argument-hint: "[no arguments — runs full sweep]"
+---
+
+# Maintenance Sweep
+
+> System-wide health check using agnix lint + 7 parallel sub-agents. Produces a consolidated report at `audits/system-audit-YYYY-MM-DD.md`. **Report-only — never modifies any files.**
+
+## When to Use
+
+- Periodic system hygiene (monthly or after major changes)
+- When the user says "maintenance sweep", "system health check", "audit my setup"
+- After adding/removing skills, hooks, agents, or rules
+- Before presenting the system to others (ensure everything is consistent)
+
+## Overview
+
+1. **Lint** — run `npx agnix .` to validate skills, hooks, agents, and CLAUDE.md
+2. **Dispatch** — launch 7 sub-agents in parallel via the Task tool
+3. **Collect** — gather each sub-agent's findings
+4. **Consolidate** — merge lint results + sub-agent findings into a single timestamped report
+5. **Present** — show key findings to the user
+
+**Python:** Always use `uv run python` or `uv pip install`. Never bare `python`, `python3`, `pip`, or `pip3`. Include this in sub-agent prompts.
+
+---
+
+## Phase 0: agnix Lint
+
+Run `npx agnix .` in the main context (no sub-agent needed — it's fast). Capture the summary line.
+
+**Pass criteria:** 0 errors. Warnings are informational only.
+
+If errors > 0, list them in the report under an **agnix Lint** section. These are structural issues in skill/hook/agent frontmatter that should be fixed.
+
+Config lives in `.agnix.toml` at project root — known false positives are already suppressed.
+
+---
+
+## Phase 1: Dispatch Sub-Agents
+
+Launch all 7 in a single message using parallel Task tool calls. Each sub-agent is `subagent_type: Explore`.
+
+**Context overflow prevention:** Instruct each sub-agent to keep its returned output concise — summary tables and key findings only (under 500 words). If detailed findings are large, the sub-agent should write them to a temp file (e.g., `/tmp/system-audit/sa-N.md`) and return only the file path + summary.
+
+All sub-agents receive shared context (Task Management root, research projects root, category directories). Full shared context block and all 6 sub-agent prompt templates:
+
+**[references/sub-agent-prompts.md](references/sub-agent-prompts.md)**
+
+Sub-agents at a glance:
+
+| # | Name | What it checks |
+|---|------|---------------|
+| 1 | Inventory Auditor | Skill/hook/agent/rule counts, symlinks, MCP alignment |
+| 2 | Bibliography & Project Hygiene | .bib files, naming, MEMORY.md presence |
+| 3 | Convention Compliance | LaTeX out/, Overleaf separation, Python env, git health |
+| 4 | Documentation Freshness | Stale counts, broken links, .context/ freshness |
+| 5 | Ecosystem Health | MCP server refs, staleness, orphans, CLI tools |
+| 6 | Skill Quality & Overlap | Bloat, staleness, cross-component overlap |
+| 7 | Santi Repo Health | Skill/rule freshness vs upstream, anonymisation, install script |
+
+---
+
+## Phase 2: Collect and Consolidate
+
+After all 7 sub-agents return, merge their findings into a single report.
+
+### Report Template
+
+Write to `audits/system-audit-YYYY-MM-DD.md`:
+
+```markdown
+# Maintenance Sweep — YYYY-MM-DD
+
+## Dashboard
+
+| Area | Status | Issues |
+|------|--------|--------|
+| agnix Lint | <OK/WARN/FAIL> | <count> |
+| Inventory | <OK/WARN/FAIL> | <count> |
+| Bibliography & Projects | <OK/WARN/FAIL> | <count> |
+| Conventions | <OK/WARN/FAIL> | <count> |
+| Documentation | <OK/WARN/FAIL> | <count> |
+| Ecosystem Health | <OK/WARN/FAIL> | <count> |
+| Skill Quality & Overlap | <OK/WARN/FAIL> | <count> |
+| Santi Repo Health | <OK/WARN/FAIL> | <count> |
+
+## agnix Lint
+<Phase 0 results: error count, warning count, any specific errors>
+
+## Inventory Audit
+<Sub-agent 1 findings>
+
+## Bibliography & Project Hygiene
+<Sub-agent 2 findings>
+
+## Convention Compliance
+<Sub-agent 3 findings>
+
+## Documentation Freshness
+<Sub-agent 4 findings>
+
+## Ecosystem Health
+<Sub-agent 5 findings>
+
+## Skill Quality & Cross-Component Overlap
+<Sub-agent 6 findings>
+
+## Santi Repo Health
+<Sub-agent 7 findings>
+
+## Recommended Actions
+<Prioritised list of things to fix, grouped by effort level>
+
+### Quick Fixes (< 5 min each)
+- ...
+
+### Medium Effort (5–30 min)
+- ...
+
+### Larger Tasks (> 30 min)
+- ...
+```
+
+---
+
+## Phase 3: Present
+
+Show the user:
+1. The dashboard table
+2. Any FAIL-status areas with specifics
+3. The quick fixes list
+4. Ask if he wants to address any issues now
+
+---
+
+## Error Handling
+
+- **Sub-agent timeout:** If a sub-agent doesn't return, note "timed out" in the report and continue with the others.
+- **Missing directories:** If a category folder is empty or doesn't exist, note "no projects found" rather than erroring.
+- **Permission issues:** If files can't be read, note "access denied" and continue.
+
+## Integration with Other Skills
+
+| Skill | Relationship |
+|-------|-------------|
+| `/bib-validate` | Run on projects flagged by the Bibliography Hygiene sub-agent |
+| `/audit-project-research` | Complements Convention Compliance with deeper per-project checks |
+| `/update-project-doc` | Fix documentation staleness found by Documentation Freshness |
+| `/sync-permissions` | Fix symlink issues found by Inventory Auditor |
+| `/atlas-review` | Full cross-system audit (local + vault + Zotero + pipeline) — deeper than this sweep |
+| `/insights-deck` | Maintenance findings can feed into system insights presentations |
+| `/audit-doc santi` | Dedicated deep audit for santi-repo — Sub-agent 7 is a quick health check; the audit skill is the full version |
+| `/sync-santi-repo` | Fix freshness issues found by Sub-agent 7 |

package/skills/system-audit/references/sub-agent-prompts.md

@@ -0,0 +1,294 @@
+# System Audit Sub-Agent Prompts
+
+> Full prompt templates for the 6 parallel sub-agents dispatched in Phase 1.
+> Referenced from `SKILL.md` — do not edit prompts here without updating the parent.
+
+## Shared Context (prepend to all prompts)
+
+```
+Task Management root: $TM/
+Global Claude config: ~/.claude/
+Project index: .context/projects/_index.md
+Research projects root: $RESEARCH_ROOT/
+Research project categories (subdirectories of the root above):
+  - Category A/
+  - Category B/
+  - Category C/
+  - Category D/
+  - Category E/
+  - Category F/
+  - Category G/
+Each category contains individual project directories.
+```
+
+## Sub-Agent 1: Inventory Auditor
+
+**Prompt:**
+```
+Audit the Task Management system inventory. Check:
+
+1. **Skill count:** Count directories in skills/ that contain a SKILL.md (exclude .DS_Store, shared/). Compare against documented count in CLAUDE.md, README.md, docs/system.md, docs/skills.md.
+2. **Hook count:** Count .sh files in hooks/ (exclude .DS_Store). Compare against documented count in CLAUDE.md, README.md, docs/system.md, docs/hooks.md.
+3. **Agent count:** Count .md files in .claude/agents/. Compare against documented count.
+4. **Rule count:** Count .md files in .claude/rules/. Compare against documented count.
+5. **Symlink health:** Verify these symlinks resolve correctly:
+   - ~/.claude/skills/ → Task Management/skills/
+   - ~/.claude/agents/ → Task Management/.claude/agents/
+   - ~/.claude/rules/ → Task Management/.claude/rules/
+   - ~/.claude/hooks/ → Task Management/hooks/
+   - ~/.claude/settings.json → Task Management/.claude/settings.json
+   - ~/.claude/CLAUDE.md → Task Management/GLOBAL-CLAUDE.md
+   - ~/.claude/statusline-command.sh → Task Management/.claude/statusline-command.sh
+6. **MCP server tool count:** The MCP server in packages/mcp-desktop/server.py registers tools as `skill-<name>` and `agent-<name>`. Count the cached skills and agents it discovers (read the discovery functions in server.py). Compare against actual skill/agent counts.
+7. **Undocumented items:** Any skills/hooks/agents/rules that exist on disk but aren't listed in their respective docs file (docs/skills.md, docs/hooks.md, docs/agents.md, docs/rules.md).
+8. **MCP server alignment:** Compare MCP servers between Claude Code (.mcp.json in project root) and Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json). Check:
+   - Servers present in both configs use the same name for the same service
+   - Servers that should be in both are not missing from either (bibliography, context7)
+   - Desktop-only servers (filesystem, skills-server) are documented
+   - No stale/removed servers remain in either config
+   - Report a side-by-side table: Server | Code | Desktop | Status
+
+Return findings as markdown with sections for each check, using checkmarks for pass and warnings for mismatches.
+```
+
+## Sub-Agent 2: Bibliography & Project Hygiene
+
+**Prompt:**
+```
+Quick scan of bibliography files and project health across the user's research projects. Check:
+
+1. **Find all .bib files** under the research projects root:
+   $RESEARCH_ROOT/
+   Categories: Category A, Category B, Category C, Category D, Category E, Category F, Category G.
+   Search each project directory and its paper/ subdirectory (2 levels deep from category).
+   Skip research/scout/ (tooling, not a research project — now lives in Task Management).
+
+2. For each .bib file found:
+   - Count entries (grep for @article, @inproceedings, @book, @misc, etc.)
+   - Check naming convention (should be references.bib or <project>.bib)
+   - Spot-check 3 keys for BBT format (AuthorYYYY-xx pattern)
+   - Flag any obvious issues (empty files, very large files >500 entries)
+
+3. **MEMORY.md presence:** For each project, check if a MEMORY.md exists in the project root. Projects that have been actively worked on should have one.
+
+4. **Summary table:** Project | Category | Bib file | Entry count | Naming OK | MEMORY.md
+
+Do NOT do a full validation — that's what /bib-validate is for. Just flag projects that should be audited in detail.
+```
+
+## Sub-Agent 3: Convention Compliance
+
+**Prompt:**
+```
+Check convention compliance across the user's research projects. Scan each project directory under:
+$RESEARCH_ROOT/
+Categories: Category A, Category B, Category C, Category D, Category E, Category F, Category G.
+Skip research/scout/ (tooling, not a research project).
+
+For each project, check:
+
+1. **LaTeX output directory:** If .tex files exist (in project root or paper/), does an out/ directory exist? Is there a .latexmkrc file?
+2. **Overleaf separation:** If a paper/ directory exists, is it a symlink? Check that paper/ contains ONLY LaTeX source files (.tex, .sty, .cls, .bst, .bbl, .bib, .latexmkrc, out/) and figures (.pdf, .png, .eps, .jpg, .svg, .tikz). Flag any code files (.py, .R, .jl, .sh, .ipynb), data files (.csv, .xlsx, .json, .dta, .parquet), or other non-LaTeX artifacts found inside paper/.
+3. **Hook executability:** All .sh files in $TM/hooks/ should be executable (chmod +x).
+4. **Python environment:** If .py files exist in the project, is there a pyproject.toml? Any sign of bare pip usage (requirements.txt without pyproject.toml, pip in scripts)?
+5. **CLAUDE.md presence:** Does each project have a CLAUDE.md?
+6. **Git health:** Is the project a git repo? Any uncommitted changes? Any untracked files that should probably be tracked?
+
+Report per-project compliance as a table:
+Project | Category | LaTeX/out | Overleaf sep. | Python env | CLAUDE.md | Git
+
+Only scan top-level project directories — don't recurse deeply into subdirectories.
+```
+
+## Sub-Agent 4: Documentation Freshness
+
+**Prompt:**
+```
+Check documentation freshness in the Task Management system at:
+$TM/
+
+Audit:
+
+1. **Stale counts in docs:** Check these files for numeric claims about skills, hooks, agents, rules, and compare against actual counts on disk:
+   - CLAUDE.md: skill count, hook count, agent count, rule count
+   - README.md: same counts
+   - docs/system.md: same counts, file tree accuracy
+   - docs/skills.md: total skill count, category counts, skill catalogue completeness
+
+2. **Broken internal links:** Check markdown links in CLAUDE.md, README.md, docs/system.md, docs/skills.md, docs/hooks.md, docs/agents.md, docs/rules.md — do the referenced files actually exist?
+
+3. **Outdated statuses in .context/:**
+   - current-focus.md: when was it last modified? (check file stat or git log)
+   - projects/_index.md: compare project list against actual directories in the research projects root. Any projects on disk but missing from the index? Any entries in the index for projects that don't exist?
+
+4. **Old session logs:** What's the most recent file in log/? How many logs exist? Any very old logs (>90 days) that could be archived?
+
+5. **Plan staleness:** Any plans in log/plans/ more than 30 days old that reference incomplete work?
+
+6. **GLOBAL-CLAUDE.md vs CLAUDE.md consistency:** Check that GLOBAL-CLAUDE.md (the slim pointer file) doesn't duplicate detailed content from CLAUDE.md. It should contain only identity, pointers, and global policies.
+
+Return as markdown with severity levels: OK, STALE, BROKEN.
+```
+
+## Sub-Agent 5: Ecosystem Health
+
+**Prompt:**
+```
+Check ecosystem health for the Task Management system at:
+$TM/
+
+Run these 4 checks:
+
+1. **MCP server health:** Find all mcp__*__ tool references in skills and agents:
+   grep -rohn "mcp__[A-Za-z0-9_-]*__[A-Za-z0-9_]*" skills/ .claude/agents/ --include="*.md" | sort -u
+
+   Extract server names (the part between mcp__ and the second __).
+
+   Then get configured MCP servers from both configs using jq:
+   cat .mcp.json | jq -r '.mcpServers // {} | keys[]' 2>/dev/null | sort -u
+   cat ~/Library/Application\ Support/Claude/claude_desktop_config.json | jq -r '.mcpServers // {} | keys[]' 2>/dev/null | sort -u
+
+   Also check for disabled servers:
+   cat .mcp.json | jq -r '.disabledMcpServers // [] | .[]' 2>/dev/null
+
+   Flag: CRITICAL if a referenced server is not configured anywhere (phantom tool).
+   Flag: WARNING if a server is configured but also in disabledMcpServers.
+   Known aliases: "claude_ai_vault" is a Claude.ai managed integration (not in local config — OK).
+   Known servers: bibliography, context7, claude_ai_vault, claude_ai_Gamma.
+
+2. **Staleness detection:** Find components not modified in 90+ days across ALL types:
+   find skills/ -name "SKILL.md" -not -path "*/shared/*" -mtime +90 -type f
+   find .claude/agents/ -name "*.md" -mtime +90 -type f
+   find hooks/ \( -name "*.sh" -o -name "*.py" \) -mtime +90 -type f
+   find .claude/rules/ -name "*.md" -mtime +90 -type f
+   find .scripts/ \( -name "*.py" -o -name "*.sh" \) -mtime +90 -type f
+
+   For each stale file, check if it's still referenced by other active components.
+   Flag: WARNING if stale AND referenced (may be outdated).
+   Flag: INFO if stale AND not referenced (candidate for archive).
+   Flag: OK if still within 90 days.
+
+3. **Orphan detection:** Find components with zero external references across ALL types.
+   For each skill directory in skills/:
+   - Read first 10 lines to check for "user-invocable: true" — if present, skip (exempt)
+   - Search for the skill name across all other skills, agents, hooks, rules, CLAUDE.md, and docs/
+   - If zero references found: flag as INFO (dead code, candidate for archive)
+   For each hook in hooks/:
+   - Check if registered in .claude/settings.json (unregistered hook = orphan)
+   - Search for the hook filename across skills, agents, docs/hooks.md
+   - If not registered AND not documented: flag as INFO
+   For each agent in .claude/agents/:
+   - Search for the agent name across skills, CLAUDE.md, docs/
+   - If zero references found: flag as INFO
+   For each rule in .claude/rules/:
+   - Search for the rule filename across CLAUDE.md, docs/, skills/
+   - Rules are auto-loaded so they're never truly orphaned, but flag if undocumented in docs/rules.md
+   For each script in .scripts/:
+   - Search for the script name across skills, hooks, CLAUDE.md, docs/
+   - If zero references found: flag as INFO
+
+4. **CLI tool availability:** Verify these tools are installed and accessible:
+   - gh (GitHub CLI)
+   - latexmk (LaTeX compilation)
+   - uv (Python package manager)
+   - jq (JSON processor)
+   - gemini (Gemini CLI for council mode)
+   - gemini (Gemini CLI for audits and council mode)
+   - node (Node.js for MCP servers)
+   For each: run "which [tool]" and "[tool] --version" (or equivalent).
+   Flag: CRITICAL if a tool referenced in CLAUDE.md as required is not installed.
+   Flag: WARNING if installed but fails to respond.
+   Flag: OK if installed and responds.
+
+Return findings as markdown with a summary table:
+| Check | OK | Warning | Critical |
+Then list each finding with severity, file path, and brief description.
+Keep output under 500 words — write details to /tmp/system-audit/sa-5.md if needed.
+```
+
+## Sub-Agent 6: Skill Quality & Cross-Component Overlap
+
+**Prompt:**
+```
+Evaluate skill quality and cross-component overlap for the Task Management system at:
+$TM/
+
+## Part A: Skill Quality
+
+For each skill directory in skills/ (excluding shared/):
+
+1. Read the SKILL.md frontmatter (name, description)
+2. Check file size (flag >300 lines as potentially bloated)
+3. Check modification date (flag >90 days as potentially stale)
+4. Spot-check for content overlap with other skills (read first 20 lines of each, look for similar descriptions or duplicate trigger phrases)
+5. Check for broken references (references to files in references/ or shared/ that don't exist)
+
+For each skill, assign a verdict:
+- **OK** — healthy, no issues
+- **BLOATED** — >300 lines, may need extraction to references/
+- **STALE** — not modified in 90+ days AND not referenced by other components
+- **OVERLAP** — significant description/trigger overlap with another skill (name it)
+- **BROKEN** — references missing files
+
+Return a summary table:
+| Skill | Lines | Last Modified | Verdict | Notes |
+
+## Part B: Cross-Component Overlap
+
+Check for functional overlap ACROSS component types (skills, hooks, agents, rules, .scripts/). Read the description/purpose of each component:
+- skills/*/SKILL.md (frontmatter description)
+- hooks/*.sh and hooks/*.py (header comment block)
+- .claude/agents/*.md (first 10 lines for purpose)
+- .claude/rules/*.md (first 10 lines for principle)
+- .scripts/*.py and .scripts/*.sh (header comment block)
+
+Flag cases where two different component types appear to do the same thing or enforce the same constraint. Common overlap patterns to check:
+- A hook gating something a rule also describes (e.g., both blocking bare python)
+- A skill doing what a .scripts/ CLI tool also does
+- An agent's workflow duplicating a skill's workflow
+- A rule describing a convention that a hook already enforces automatically
+
+For each overlap found, assess:
+- **Complementary** — both are needed (e.g., hook enforces at runtime, rule instructs at planning time)
+- **Redundant** — one could be removed without loss
+- **Conflicting** — they give contradictory instructions
+
+Return a cross-component table:
+| Component A | Component B | Type | Assessment | Notes |
+
+Only flag genuine overlaps — a rule mentioning "use uv" and a hook enforcing "use uv" is complementary (expected), not redundant. Focus on cases where work is truly duplicated or instructions conflict.
+
+Keep total output under 500 words. Write details to /tmp/system-audit/sa-6.md if needed.
+```
+
+## Sub-Agent 7: Santi Repo Health
+
+**Prompt:**
+```
+Quick health check of Santiago's starter kit at:
+$TM/public/santi-repo/
+
+This is a downstream copy of skills/rules for Santiago. Check:
+
+1. **Skill count:** Count directories in santi-repo/skills/ that contain a SKILL.md. Compare against the count stated in README.md and ADAPTATION-NEEDED.md.
+
+2. **Rule count:** Count .md files in santi-repo/rules/. Compare against the count stated in README.md.
+
+3. **Skill freshness (sample):** Pick 5 random skills from santi-repo/skills/ and diff their SKILL.md against the upstream version in $TM/skills/. Flag any with substantive upstream changes not reflected in santi-repo (ignoring anonymisation differences like "the user" → "the user").
+
+4. **Rule freshness:** For each rule in santi-repo/rules/, compare against the upstream version in $TM/.claude/rules/. Flag substantive differences (ignoring anonymisation).
+
+5. **Anonymisation spot-check:** Grep all .md files in santi-repo/ for leaked personal details:
+   - "the user" (case-sensitive, excluding "Made by the user" in README.md)
+   - "$HOME" (macOS path)
+   - "Warwick", "Southampton", "Bath", "LSE" (institution names)
+   Exclude GitHub URLs containing "user" (intentional).
+
+6. **README accuracy:** Does the skill count in README.md match the actual count? Does the rule count match?
+
+7. **Install script:** Does install.sh exist and contain no macOS-specific commands (sed -i '', open, brew without apt)?
+
+Return a summary table:
+| Check | Status | Notes |
+Then list any issues found. Keep output under 300 words.
+```

package/skills/task-management/SKILL.md

@@ -0,0 +1,131 @@
+---
+name: task-management
+description: "Use when you need help with daily planning, weekly reviews, meeting actions, or vault task queries."
+allowed-tools: Read, Write, Edit, AskUserQuestion
+---
+
+# the user's Task Management System
+
+## MCP Pre-Check
+
+Before any vault-dependent workflow, probe taskflow MCP availability with a lightweight search. If unavailable, skip vault queries and offer local-only fallbacks per [`shared/mcp-degradation.md`](../shared/mcp-degradation.md).
+
+## System Overview
+
+This is a hybrid vault + local context library system. Before taking action:
+
+1. **Read context files** in `.context/` to understand current state
+2. **Query vault** for dynamic task data
+3. **Ask questions** before dumping lists (the user prefers this)
+4. **Update context** after sessions
+
+## Key Locations
+
+### Context Library (`.context/`)
+- `profile.md` — Who the user is, roles, research areas
+- `current-focus.md` — What they're working on NOW (update this!)
+- `projects/_index.md` — All active projects
+- `workflows/` — How to help with specific tasks
+- `preferences/` — Priority definitions, naming conventions
+- `people/` — Supervisors and collaborators
+
+### Vaults
+- **Tasks Tracker**: `collection://YOUR-TASKS-DATABASE-ID-HERE`
+- **Research Pipeline**: `collection://YOUR-PIPELINE-DATABASE-ID-HERE`
+
+## Workflows
+
+### Daily Planning
+
+When the user asks to plan their day:
+
+1. **Read** `.context/current-focus.md` and `.context/workflows/daily-review.md`
+2. **Query vault** for:
+   - Overdue tasks (Due date < today, Status != Done)
+   - Due today (Due date = today)
+   - High priority items
+3. **Ask orientation questions**:
+   - "How's your energy today?"
+   - "Any fixed commitments?"
+   - "What were you working on yesterday?"
+4. **Help prioritise** based on answers
+5. **Update** `current-focus.md` at end of session
+
+### Meeting Action Extraction
+
+When the user asks to extract actions from a meeting:
+
+1. **Read** `.context/workflows/meeting-actions.md`
+2. **Find the meeting transcript** in vault (pages starting with @Date)
+3. **Extract action items** looking for:
+   - "I'll...", "I need to...", "I should..."
+   - "Can you...", "Please..."
+   - "We agreed to...", "Next step is..."
+4. **Create tasks in vault** with full context:
+   - Task name (action verb + object)
+   - Project (infer from context)
+   - Source: "Meeting"
+   - Due date (if mentioned)
+   - Description (meeting context)
+
+### Weekly Review
+
+When the user asks for weekly review:
+
+1. **Read** `.context/workflows/weekly-review.md`
+2. **Query vault** for:
+   - Completed tasks this week
+   - Overdue tasks
+   - Upcoming deadlines
+3. **Guide through reflection**:
+   - What got done?
+   - What didn't happen?
+   - What emerged?
+4. **Help plan next week's Big 3**
+5. **Update** `current-focus.md` and `projects/_index.md`
+
+### Task Creation
+
+When creating tasks in vault:
+
+**Required fields:**
+- Task name (action verb + specific object)
+- Status: "Not started"
+
+**Recommended fields:**
+- Project: Match to existing project
+- Source: Meeting, Email, Supervisor request, Self-initiated, etc.
+- Priority: High/Medium/Low (see `.context/preferences/priorities.md`)
+- Due date: If known
+- Task type: 📝 Writing, 📚 Reading, 🔬 Research, 📅 Meeting, 📋 Admin, 📧 Communication
+
+### Research Pipeline
+
+For paper-related queries:
+
+1. **Query Research Pipeline database** for paper status
+2. **Stages**: Idea → Literature Review → Drafting → Submitted → R&R → Published
+3. **Link tasks to papers** via Project property
+
+## the user's Preferences
+
+- **Questions over lists** — Don't dump task lists, ask what to focus on
+- **Full context for actions** — Include who, what, why, when in task descriptions
+- **Flexible/reactive style** — They work on what feels right, help them navigate
+- **Daily review preferred** — Short check-ins work better than big weekly sessions
+- **multiple active projects** — Help manage cognitive load, don't add more
+
+## Important Context
+
+- PhD Year 1 at [University]
+- Teaching
+- 4 supervisors across institutions
+- Research: [your research areas]
+- Current papers: journal revision, [Project] theory, [Project]
+
+## After Every Session
+
+Update `.context/current-focus.md` with:
+- What was worked on
+- Where things were left off
+- What's next