Split Epiphany route authority
This commit is contained in:
parent
98df7bf3d2
commit
996eeed371
7 changed files with 1072 additions and 1063 deletions
|
|
@ -435,15 +435,20 @@ processor to about 18,362 lines. The eleventh cut moved the Epiphany JSON-RPC
|
|||
route-handler cluster into the child module
|
||||
`vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_routes.rs`,
|
||||
taking the processor to about 16,088 lines. This route child module still uses
|
||||
parent-internal visibility; it is a staging wound, not final purity.
|
||||
parent-internal visibility; it is a staging wound, not final purity. The twelfth
|
||||
cut split that route chamber into read/proposal routes at
|
||||
`vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_read_routes.rs`
|
||||
and mutation/launch/accept/index routes at
|
||||
`vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_mutation_routes.rs`.
|
||||
The processor is about 16,089 lines after the split; this pass clarified
|
||||
authority rather than removing another large block.
|
||||
|
||||
Also: MCP itself is allowed to be JSON. The target is not "replace MCP JSON";
|
||||
the target is an Epiphany-owned boundary that speaks typed Epiphany
|
||||
intent/result/receipt documents internally and normal MCP JSON-RPC externally.
|
||||
|
||||
Continue with the actual whale-carcass cut: split the Epiphany route child
|
||||
module into read-only routes versus mutation/launch/accept routes, then move
|
||||
mutation policy and runtime authority out of
|
||||
Continue with the actual whale-carcass cut: move mutation policy and runtime
|
||||
authority out of the mutation route module and out of
|
||||
`codex_message_processor.rs` into Epiphany-owned app-server modules, moving
|
||||
logic further into `epiphany-core` whenever it owns real policy. Success is a
|
||||
visibly smaller `codex_message_processor.rs`, not merely nicer payload names. Do
|
||||
|
|
|
|||
|
|
@ -62,6 +62,6 @@ Rule: JSON is allowed at external schema/wire boundaries. Inside Epiphany, live
|
|||
3. Done: replace role `statePatch` JSON with typed map/planning/graph patch documents. `EpiphanyRoleFindingInterpretation` now carries `EpiphanyRoleStatePatchDocument`; app-server maps it to the legacy protocol patch without JSON round-tripping.
|
||||
4. Done: replace runtime job result JSON projections. Runtime-spine results now project through typed role/reorient interpreters instead of `runtime_job_result_to_role_json` and `runtime_job_result_to_reorient_json`.
|
||||
5. Done: replace app-server launch protocol `input_json` / `output_schema_json` with typed launch documents and `output_contract_id`.
|
||||
6. Continue the whale-carcass cut: the first cuts moved Epiphany launch doctrine and protocol-to-core launch document mapping into `epiphany_launch.rs`, role/reorient result projection into `epiphany_results.rs`, scene projection into `epiphany_scene.rs`, freshness/reorientation mapping into `epiphany_reorient.rs`, context/planning/graph-query projection into `epiphany_context.rs`, jobs projection into `epiphany_jobs.rs`, retrieve projection into `epiphany_retrieve.rs`, pressure/pre-compaction checkpoint projection into `epiphany_pressure.rs`, CRRC/coordinator/role-board projection plus acceptance/evidence signal helpers into `epiphany_coordinator.rs`, and the Epiphany JSON-RPC route-handler cluster into `codex_message_processor/epiphany_routes.rs`, reducing `codex_message_processor.rs` from about 21,263 to 16,088 lines. Next split read-only routes from mutation/launch/accept routes and move mutation policy toward native CultNet/CultCache contracts; success means the processor keeps shrinking and ownership gets clearer, not merely better typed.
|
||||
6. Continue the whale-carcass cut: the first cuts moved Epiphany launch doctrine and protocol-to-core launch document mapping into `epiphany_launch.rs`, role/reorient result projection into `epiphany_results.rs`, scene projection into `epiphany_scene.rs`, freshness/reorientation mapping into `epiphany_reorient.rs`, context/planning/graph-query projection into `epiphany_context.rs`, jobs projection into `epiphany_jobs.rs`, retrieve projection into `epiphany_retrieve.rs`, pressure/pre-compaction checkpoint projection into `epiphany_pressure.rs`, CRRC/coordinator/role-board projection plus acceptance/evidence signal helpers into `epiphany_coordinator.rs`, and Epiphany JSON-RPC route handling under `codex_message_processor/` split into read/proposal routes and mutation/launch/accept/index routes, reducing `codex_message_processor.rs` from about 21,263 to 16,089 lines. Next move mutation policy and runtime authority toward native CultNet/CultCache contracts; success means the processor keeps shrinking and ownership gets clearer, not merely better typed.
|
||||
7. Seal MCP payloads behind an Epiphany boundary that speaks typed Epiphany documents internally and MCP JSON externally.
|
||||
8. Audit heartbeat cognition values and either type them or explicitly expire them.
|
||||
|
|
|
|||
Binary file not shown.
|
|
@ -278,5 +278,5 @@ rejected_paths:
|
|||
reason: making Epiphany cleaner inside Codex still preserves the wrong organ; Codex must be starved down to OpenAI subscription auth/model transport, while Epiphany data and subsystem communication move to typed CultCache/CultNet.
|
||||
|
||||
current_status:
|
||||
summary: The Codex starvation pass has five typing receipts plus eleven real carcass cuts: Epiphany launch doctrine/prompt/builders and protocol-to-core launch document mapping moved into `vendor/codex/codex-rs/app-server/src/epiphany_launch.rs`, role/reorient result projection/note rendering moved into `vendor/codex/codex-rs/app-server/src/epiphany_results.rs`, scene projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_scene.rs`, freshness/reorientation input mapping moved into `vendor/codex/codex-rs/app-server/src/epiphany_reorient.rs`, context/planning/graph-query projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_context.rs`, jobs projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_jobs.rs`, retrieve projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_retrieve.rs`, pressure/pre-compaction checkpoint projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_pressure.rs`, CRRC/coordinator/role-board projection plus acceptance/evidence signal helpers moved into `vendor/codex/codex-rs/app-server/src/epiphany_coordinator.rs`, and the Epiphany JSON-RPC route-handler cluster moved into child module `vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_routes.rs`. The processor is down from about 21,263 lines to 16,088 lines. This is progress, not absolution: the route module still has parent-internal visibility and accept/launch mutations still live in the Codex adapter layer. MCP itself is correctly JSON at its protocol edge; the target is typed Epiphany intent/result/receipt wrappers around MCP, not replacing MCP JSON.
|
||||
next_action: Continue the actual whale-carcass cut: split the child Epiphany route module into read-only surfaces versus mutation/launch/accept surfaces, then push mutation policy further into `epiphany-core` or native CultNet/CultCache runtime contracts. Success metric remains a visibly smaller processor and clearer ownership, not merely typed payloads.
|
||||
summary: The Codex starvation pass has five typing receipts plus twelve real ownership cuts: Epiphany launch doctrine/prompt/builders and protocol-to-core launch document mapping moved into `vendor/codex/codex-rs/app-server/src/epiphany_launch.rs`, role/reorient result projection/note rendering moved into `vendor/codex/codex-rs/app-server/src/epiphany_results.rs`, scene projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_scene.rs`, freshness/reorientation input mapping moved into `vendor/codex/codex-rs/app-server/src/epiphany_reorient.rs`, context/planning/graph-query projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_context.rs`, jobs projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_jobs.rs`, retrieve projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_retrieve.rs`, pressure/pre-compaction checkpoint projection moved into `vendor/codex/codex-rs/app-server/src/epiphany_pressure.rs`, CRRC/coordinator/role-board projection plus acceptance/evidence signal helpers moved into `vendor/codex/codex-rs/app-server/src/epiphany_coordinator.rs`, the Epiphany JSON-RPC route-handler cluster moved under `vendor/codex/codex-rs/app-server/src/codex_message_processor/`, and that cluster is now split into read/proposal routes (`epiphany_read_routes.rs`) versus mutation/launch/accept/index routes (`epiphany_mutation_routes.rs`). The processor is down from about 21,263 lines to 16,089 lines. This is progress, not absolution: the route modules still have parent-internal visibility and mutation authority still lives in the Codex adapter layer. MCP itself is correctly JSON at its protocol edge; the target is typed Epiphany intent/result/receipt wrappers around MCP, not replacing MCP JSON.
|
||||
next_action: Continue the actual whale-carcass cut: move mutation policy and runtime authority out of the mutation route module into `epiphany-core` or native CultNet/CultCache runtime contracts, starting with the launch/accept/update helper duplication. Success metric remains a visibly smaller processor and clearer ownership, not merely typed payloads.
|
||||
|
|
|
|||
|
|
@ -1,4 +1,5 @@
|
|||
mod epiphany_routes;
|
||||
mod epiphany_mutation_routes;
|
||||
mod epiphany_read_routes;
|
||||
|
||||
use crate::bespoke_event_handling::apply_bespoke_event_handling;
|
||||
use crate::bespoke_event_handling::maybe_emit_hook_prompt_item_completed;
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load diff
1058
vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_read_routes.rs
vendored
Normal file
1058
vendor/codex/codex-rs/app-server/src/codex_message_processor/epiphany_read_routes.rs
vendored
Normal file
File diff suppressed because it is too large
Load diff
Reference in a new issue