1 evidence-backed claim.
claude-code-2026-08-10-docs →MECHANISM · EXTENSIBILITY · CORE
Extension Runtime
Load and execute hooks, plugins, skills, custom tools, or provider adapters without collapsing extension code into the core loop.
3 lessons6 agent snapshots0 experiments
extensibilityextension-runtimeproblem → policy → evidence
L0Intuitionpartial
L1Buildmissing
L2Engineeringpartial
L3Architecturepartial
L4Researchpartial
DEPENDENCY GRAPH · READER PATH
What feeds it, and what does it unlock?
Arrows show learning and design dependencies, not runtime data flow. Follow any node to continue through the Atlas.
extension-runtimeExtension RuntimeCurrent research boundary
UnlocksNo dependents yet
Load and execute hooks, plugins, skills, custom tools, or provider adapters without collapsing extension code into the core loop.
The structured record is in the Atlas, but the L0–L4 article has not passed content review.
AGENT MAPPING · EVIDENCE ONLY
Snapshot implementations
An implementation enters the map only when both a snapshot and claims exist; unknowns remain visible.
1 evidence-backed claim.
codex-2026-08-10-source →1 evidence-backed claim.
grok-build-2026-08-10-source →1 evidence-backed claim.
opencode-2026-08-10-source →1 evidence-backed claim.
pi-2026-08-10-source →1 evidence-backed claim.
reasonix-2026-08-10-source →No implementation currently clears the Snapshot + Claim threshold; this is not a claim that the product lacks the capability.
Inspect coverage boundary →CLAIMS · EVIDENCE LEDGER6 RECORDS
official-docreviewed
claude-code-doc-hooks
Claude Code's official hook contract exposes lifecycle events including pre- and post-tool use, permission requests, session start/end, and stop; PreToolUse hooks can deny a call before normal permission-mode checks, while allow decisions cannot override stricter deny rules.
- Agent
- Claude Code
- Snapshot
- claude-code-2026-08-10-docs
- Reviewed
- 2026-08-10
source + sourcereviewed
codex-source-extension-runtime
At the pinned revision, Codex assembles selected skills and plugins into turn context and runs typed lifecycle hooks around session start, tool use, permissions, stop, compaction, and session end.
- Agent
- Codex
- Snapshot
- codex-2026-08-10-source
- Reviewed
- 2026-08-10
source + sourcereviewed
grok-build-source-extension-runtime
Grok Build's pinned extension surface includes plugins, prompt skills, MCP tools, and lifecycle hooks, with pre-tool gates able to modify or block operations.
- Agent
- Grok Build
- Snapshot
- grok-build-2026-08-10-source
- Reviewed
- 2026-08-10
source + source + sourcereviewed
opencode-source-extension-runtime
OpenCode's pinned source discovers local skills, loads configured plugins, and imports MCP tools into the common tool-resolution boundary, with trust and permission still configuration-dependent.
- Agent
- OpenCode
- Snapshot
- opencode-2026-08-10-source
- Reviewed
- 2026-08-10
source + sourcereviewed
pi-source-extension-runtime
Pi's coding-agent layer uses an extension runner to dispatch lifecycle events and lets extensions add or replace tools, commands, UI, providers, context transforms, and compaction behavior without modifying the low-level agent loop.
- Agent
- Pi
- Snapshot
- pi-2026-08-10-source
- Reviewed
- 2026-08-10
source + sourcereviewed
reasonix-source-extension-runtime
Reasonix plugin manifests can contribute hooks and MCP servers, while lazy MCP placeholders keep cached tool metadata stable and defer process startup or live schema reconciliation until needed.
- Agent
- Reasonix
- Snapshot
- reasonix-2026-08-10-source
- Reviewed
- 2026-08-10
EXPERIMENTS · EXERCISES
What has actually been tested?
Formal experiments are separate from course exercises. Exercises can validate the reference implementation but cannot replace Native evidence from a real agent.
Formal experiment · not registered
No experiment record means untested, not passed. The course checks below cover only the deterministic reference harness.
s03-tool-dispatchobserve
Verify the tool roundtripThe request, result, and second model request appear in causal order.
python3 -m curriculum.golden verify s03-tool-dispatchs03-tool-dispatchmodify
Test the dispatch contractThe tool result returns to the next turn as a tool message bound to tool_call_id.
python3 -m unittest curriculum.tests.test_vertical_slice.VerticalSliceTests.test_s03_tool_roundtrip -vs05-instructionsobserve
Verify instruction inputThe run matches the committed trace without leaking a local absolute path.
python3 -m curriculum.golden verify s05-instructionss05-instructionsmodify
Test root-to-leaf precedenceRoot instructions remain before package instructions and an external cwd is rejected.
python3 -m unittest curriculum.tests.test_vertical_slice.VerticalSliceTests.test_s05_root_to_leaf_instructions curriculum.tests.test_vertical_slice.VerticalSliceTests.test_instruction_discovery_rejects_external_cwd -vs09-memory-skillsobserve
Verify on-demand loadingmemory.read and skill.load each appear once with source or skill identity.
python3 -m curriculum.golden verify s09-memory-skillss09-memory-skillsmodify
Test the retrieval contractQueries return only relevant source-attributed memory; unknown skills fail explicitly rather than being silently injected.
python3 -m unittest curriculum.tests.test_vertical_slice.VerticalSliceTests.test_s09_retrieves_memory_and_loads_skill -v01
Which extension capabilities require an explicit trust decision?
02
How should hook conflicts and non-deterministic completion order be surfaced?
03
Can one portable extension API preserve each harness's safety model?