Garden

Tend the personal knowledge base incrementally. Small regular attention beats massive occasional cleanups.

Gardening Activities

Modes

lint [area]

Validate frontmatter structure and YAML validity. Uses scripts/lint_frontmatter.py.

uv run python $AOPS/aops-tools/skills/garden/scripts/lint_frontmatter.py <path> [--recursive] [--fix] [--errors-only]

What it checks:

Common YAML issues requiring manual fix:

• •title: [Learn] something - brackets parsed as YAML array; quote the title
• •aliases: [x] followed by - x - conflicting YAML syntax
• •Wikilinks in values - [[link|alias]] can break YAML parsing

Usage:

# Scan for issues
uv run python lint_frontmatter.py data/tasks/ --recursive --errors-only

# Fix delimiter issues automatically
uv run python lint_frontmatter.py data/tasks/inbox/ --fix

scan [area]

Health check. Count orphans, broken links, stale content, sparse notes, duplicates, orphan implementation docs.

Orphan Implementation Doc Detection: Files in experiments/ that describe features with existing specs should be synthesized, not left as separate files. Scan identifies these:

1. •List files in $AOPS/experiments/*.md
2. •For each, check title/content against specs/ filenames and content
3. •If match found → report as "orphan implementation doc - synthesize into [spec]"
4. •Output: "Found N implementation docs that should be merged into specs"

Note: Per AXIOM #28, episodic observations go to tasks, not local files.

weed [area]

Fix broken [[wikilinks]], remove dead references, flag duplicates.

prune [area]

Archive stale sessions (>30 days), compress verbose notes, update temporal metadata.

link [area]

Connect orphans and enforce semantic link density per [[HEURISTICS.md#H34]].

What to check:

1. •Zero-backlink orphans: Files with no incoming links
2. •Semantic orphans: Files in same folder with overlapping topics but no mutual links
3. •Hub disconnection: Project hubs that don't link to their key content files
4. •Missing bidirectional links: If A references B's topic, both should link

Detection process:

1. •List files in target area
2. •For each project folder, check: does hub link to substantive content (not just [[meetings]])?
3. •For files with overlapping tags in same folder, check: do they link to each other?
4. •Use mcp__memory__retrieve_memory on file titles - if memory returns "related" files that don't link, flag them

Fix process:

• •Add wikilinks IN PROSE where semantic relationships exist (per [[HEURISTICS.md#H7b]] - no "see also" sections)
• •Ensure hub files link to key strategic/analysis documents
• •Ensure related analyses link bidirectionally

cultivate [area]

Add missing frontmatter, expand sparse notes, improve titles for searchability.

consolidate [topic]

Merge notes on same topic, extract patterns from logs into learning files.

map [folder]

Create missing folder/folder.md MoCs, update stale MoCs.

dry [area]

Find restated content that should be links. See DRY Enforcement below.

synthesize [area]

De-temporalize content: strip deliberation artifacts from specs, and delete/consolidate agent-generated temporal logs. See [[HEURISTICS.md#H23]] and [[AXIOMS.md#13]] (Trust Version Control).

Spec Index Maintenance: After synthesizing any spec, ensure specs/specs.md is updated:

1. •Check if spec is listed in the index
2. •If not listed → add to appropriate category section
3. •If status changed → update the index entry
4. •Commit spec + index together

What's SACRED (never touch):

• •User-written meeting notes, file notes, journal entries
• •Research records, annotations, observations
• •Anything the user authored as a primary document

What can be de-temporalized (agent-generated cruft):

• •Event logs: "X Cleanup - Month YYYY.md" where git has the commits
• •Decision records after implementation: options/alternatives sections
• •Temporal markers in reference docs: "as of YYYY", "TODO:", "considering"

Detection patterns:

• •Files named with dates/months that document completed work
• •Specs with status: Implemented containing ## Options or ## Alternatives
• •Reference docs with temporal language that implies incompleteness
• •Episodic content that should be tasks, not local files

Workflow for specs:

1. •Scan for implemented items with deliberation cruft
2. •For each: "Archive rationale to decisions/? Strip from spec?"
3. •Update spec, preserving acceptance criteria and current behavior

Workflow for temporal logs:

1. •Check: Is this user-written or agent-generated?
2. •If agent-generated event record: Does git have the commits?
3. •Extract any reusable heuristics → promote to HEURISTICS.md
4. •Delete the temporal file (git is the archive)

Does NOT delete user content - only agent-generated temporal records whose value is now in version control.

DRY Enforcement (Critical)

Problem: Content copied from core docs (AXIOMS, ACCOMMODATIONS, HEURISTICS) into other files instead of linking. Creates maintenance burden and drift.

Detect restated content:

1. •Look for sections labeled "from [[X]]" or "per [[X]]" - these restate instead of link
2. •Find multi-line quotes from core framework docs
3. •Detect duplicate bullet points across files

Example violation (from a spec file):

**Communication Guidelines** (from [[ACCOMMODATIONS.md]]):

- Match user's preparation level...
- Proactive action: Don't ask permission...

Fix: Remove the restated content entirely. The framework already loads ACCOMMODATIONS.md via hooks - no need to repeat or even link unless critically specific to this context.

Rule: Core docs (AXIOMS, ACCOMMODATIONS, HEURISTICS, RULES) apply everywhere implicitly. Don't restate. Don't even link unless the specific section is uniquely relevant.

Workflow

1. •Ask what area to tend (or pick based on recent activity)
2. •Scan to assess health
3. •Pick mode based on findings
4. •Work small batches - 3-5 notes at a time
5. •Surface decisions - confirm before deletions
6. •Commit frequently - logical chunks

Session length: 15-30 minutes max. Gardening is sustainable when light.

Area Targeting

Default: highest-activity areas (recent modifications).

Health Metrics

Anti-Patterns

• •Marathon cleanup sessions → Small, frequent instead
• •Delete without extracting value → Preserve then prune
• •Reorganize folder structures → Use MoCs for navigation
• •Restate core docs → Link or omit entirely
• •Perfectionism → Progress over perfection
• •## Relations or 'references' boilerplate sections → Remove entirely

Named Session Logs Anti-Pattern

Problem: Individual session files like "Framework Logger Agent - First Production Use.md" or "ZotMCP Implementation Session 2025-11-22.md" duplicating what daily logs capture.

Why it's wrong:

1. •Daily logs are the authoritative record of what happened each day
2. •Named session files fragment history across multiple locations
3. •Creates maintenance burden and stale cross-references
4. •Reusable patterns go to HEURISTICS.md or tasks, not session logs

What to keep:

• •Daily logs: YYYYMMDD-daily.md
• •Raw transcripts: claude/*.md (machine-generated, useful for analysis)

What to delete:

• •Named session files: "X Implementation Session.md", "Y - First Use.md"
• •Session summaries that restate daily log content
• •Full transcript copies outside claude/ subdirectory

If valuable patterns exist: Extract to HEURISTICS.md or create a task via mcp__plugin_aops-core_tasks__create_task(), then delete the session log.