Framework Context
Overview
Provide a consistent, source-anchored model of this framework so agents can reason about system-level behavior without guessing.
When to Use
System-level reflection or metacognitive assessment
Planning tasks that depend on routing/memory/workflow architecture
Capability-gap analysis before recommending ecosystem evolution
Iron Laws
ALWAYS
load framework context before any system-level reflection, planning, or capability-gap analysis — reasoning about architecture without grounding produces hallucinated paths and phantom agents.
NEVER
fabricate file paths or agent names that are not confirmed by reading canonical sources — invented paths break downstream agents that try to use them.
ALWAYS
scope the context load to only what the consuming task needs (
memory
,
agents
,
workflows
,
hooks
, or
all
) — loading all sources for a narrow task wastes tokens and buries relevant signals.
NEVER
write or modify framework files from within this skill — framework-context is read-only; mutations require the appropriate creator/updater skill.
ALWAYS
report missing sources explicitly as
missing source:
rather than silently omitting a section — silent omissions cause downstream agents to make decisions on incomplete context.
Anti-Patterns
Anti-Pattern
Why It Fails
Correct Approach
Skipping context load before reflection
Reflection uses stale or hallucinated routing/memory assumptions
Always invoke framework-context first; never reflect from memory alone
Loading full
all
scope for a narrow task
Token budget consumed by irrelevant sections; key signal buried
Pass
--scope memory
or
--scope agents
to limit output to what's needed
Inferring file paths from naming conventions
Paths change; inferred paths break agent pipelines
Always read canonical sources and report actual paths found
Writing to framework files inside this skill
Bypasses creator workflow and post-creation integration steps
Use appropriate creator/updater skill for any write operations
Silently omitting sections when source is missing
Consumer assumes context is complete; makes decisions on gaps
Report
missing source:
explicitly for every unresolvable section
Step 0: Resolve Scope
Accept
scope
argument:
memory | agents | workflows | hooks | all
Default to
all
Step 1: Load Canonical Sources
Read only what is required by scope:
Memory:
.claude/docs/MEMORY_SYSTEM.md
Agent registry/routing:
.claude/context/agent-registry.json
,
.claude/lib/routing/routing-table.cjs
Reflection flow hooks:
.claude/hooks/reflection/reflection-queue-processor.cjs
,
.claude/hooks/reflection/reflection-step0-guard.cjs
Workflow catalog:
.claude/docs/@ENTERPRISE_WORKFLOWS.md
Global framework references:
.claude/CLAUDE.md
Step 2: Emit Structured Context
Output sections in this exact order:
Memory System
Agents and Routing
Workflows
Hooks
Directory Layout
Each section must include:
2-6 concise bullets
At least one concrete path reference
Behavior notes (what triggers what)
Step 3: Scope Filter
If
scope != all
, return only relevant section(s)
Never fabricate unknown paths or flows
If a source is missing, state
missing source:
Step 3: Output
Return markdown only; do not write framework files from this skill.
// Full framework model
Skill
(
{
skill
:
'framework-context'
}
)
;
// Memory-only context for reflection prep
Skill
(
{
skill
:
'framework-context'
,
args
:
'--scope memory'
}
)
;
// Workflow/hook-only context for integration analysis
Skill
(
{
skill
:
'framework-context'
,
args
:
'--scope workflows'
}
)
;
Memory Protocol (MANDATORY)
Before starting:
Get-Content
.
claude/context/memory/learnings
.
md
-
TotalCount 120
After completing:
New framework-context pattern ->
.claude/context/memory/learnings.md
Broken/ambiguous framework mapping ->
.claude/context/memory/issues.md
Architectural interpretation decision ->
.claude/context/memory/decisions.md
ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.