all

All (/all)

Purpose

Single entry point to search EVERYTHING - your local data (chat history, knowledge base, code, docs) AND the web (current documentation, best practices, Stack Overflow). Results are merged and ranked by relevance, then intelligently filtered via three-layer architecture.

When to Use

Command	Searches	Speed	When To Use
/find	Local data only	<1s	"What did we discuss?"
/web	Web only	5-10s	"What's the current best practice?"
/all	Both + merged results	1-10s	"I want to see everything"

Three-Layer Filtering Architecture

Layer	Responsibility	When Applied
Layer 1A	Volume control, deduplication, quality floor	Always
Layer 1B	Semantic clustering (Jaccard similarity)	Always
Layer 1C	Query complexity scoring	Always
Layer 1D	Adaptive result limits (token-aware)	Always
Layer 2	Semantic relevance via Agent tool (LLM-based)	Conditional (auto-triggers)
Layer 3	Presentation formatting	Always

Layer 2 Auto-triggers when:

• Result count > 20 (configurable: --context-threshold N)
• Query contains context hints ("we discussed", "for the X feature")
• Query complexity score high (>60) + sufficient results (>15)

User overrides:

• --no-context-filter: Skip Layer 2
• --force-context-filter: Force Layer 2 even for small result sets
• --context-threshold N: Adjust trigger threshold (default: 20)

Execution Model

Run the orchestrator as a module from the plugin root:

python -m skills.all.orchestration "query" [--mode auto --limit 30 --no-context-filter]

• Layer 1 (fetch, dedup, semantic clustering, adaptive limits) and Layer 3 (formatting) are deterministic Python.
• Layer 2 is keyword-based relevance filtering. The LLM/Agent rerank sketched in agent_filter.py was never wired — it always falls back to keywords. For a true semantic rerank on large result sets, apply the Agent tool to the script's output yourself.

Quick Usage

/all "python async patterns"                          # Auto-filtering (default)
/all "what did we decide about auth"                   # Context-aware (auto-triggers Layer 2)
/all "best practices" --mode unified                   # Force local + web
/all "what did we discuss" --mode local-only           # Fast, no web APIs
/all "query" --no-context-filter                       # Layer 1 only
/all "microservices patterns" --force-context-filter   # Always apply Layer 2

More examples and advanced options: See references/usage-examples.md

Output Format

Results display as [score] SOURCE: title with preview text. When Layer 2 activates, results are grouped by theme with key insights extracted per group.

Full output format examples: See references/output-format.md

Validation Rules

• Verify actual web API calls were made before citing results
• Ensure URL and content are real before citing web sources
• Always indicate local vs web source attribution
• Report empty results clearly for both sources
• Web search failures fall back to local results with warning

Performance

Mode	Local	Web	Total	Use Case
auto	<1s	0-10s	0-11s	Default, adaptive
unified	<1s	5-10s	5-11s	Comprehensive
local-only	<1s	0s	<1s	Fast, private
web-fallback	<1s	0-10s	0-11s	Quality-focused

Layer 1 total: <1 second. Layer 2: <5 seconds when triggered.

Reference Files

File	Contents
orchestration.py	Three-layer search orchestrator (python -m skills.all.orchestration)
references/implementation-details.md	Layer behavior, trigger conditions, error handling
references/usage-examples.md	Usage examples and advanced CLI options
references/output-format.md	Output format examples (Layer 1 vs Layer 2)
references/migration-notes.md	Migration notes (older /find+/web workflow → /find+/web+/all)
references/troubleshooting.md	Common issues and solutions
references/performance-tuning.md	Speed vs coverage tuning guide