feat(apex): bootstrap ApexCode safety and evaluation foundation by kratosboom · Pull Request #1 · GPRO-Master/ApexCode

kratosboom · 2026-08-22T05:46:36Z

Summary

Bootstrap ApexCode as an upstream-friendly, independent extension of OpenAI Codex.

This PR intentionally avoids changing the upstream Codex runtime. It establishes the first isolated ApexCode implementation slice and public project identity.

What changes

identify ApexCode clearly in README.md while retaining the upstream Codex quickstart/content
extend SECURITY.md with separate upstream-vs-ApexCode reporting guidance and safety invariants
add dependency-free apexcode/apex-policy Rust crate
define R0-R5 risk levels and default policy decisions
fail closed for unknown operation risk
add five focused unit tests for the initial policy mapping
add benchmarks/APEXBENCH.md for reproducible high-complexity evaluation

Already established on main

During bootstrap, three additive documentation files were written directly to main before branch targeting was corrected:

UPSTREAM.md
ARCHITECTURE.md
ROADMAP.md

No Codex runtime code was changed by those direct documentation commits. History was not rewritten or force-reset. This PR is based on that documented main state.

Safety

no production infrastructure changes
no secrets or proprietary GPRO source included
no Codex runtime behavior changed
no dependency added to the upstream Codex workspace
no merge requested as part of this bootstrap

Validation

The apex-policy crate is dependency-free and includes five unit tests.

Local execution was not completed because the available execution environment has neither outbound GitHub DNS access nor a Rust/Cargo toolchain. Tests must therefore be treated as UNVERIFIED until CI or a Rust-enabled environment executes them.

Suggested focused verification:

cargo fmt --manifest-path apexcode/apex-policy/Cargo.toml -- --check
cargo test --manifest-path apexcode/apex-policy/Cargo.toml

Review gate

Keep this PR in draft until:

focused Rust formatting/test evidence is green,
licensing/provenance is reviewed,
security wording is reviewed,
the isolation strategy is accepted.

Do not merge automatically.

ApexCode foundation update

Implemented:

apex-policy
apex-task-state
apex-evidence

Verification:

cargo fmt --manifest-path apexcode/Cargo.toml -- --check — Cargo does not support target discovery for this virtual workspace without --all; the equivalent new-crate rustfmt --edition 2024 --check passed.
cargo test --manifest-path apexcode/Cargo.toml --workspace — PASS with the local GNU-linker workaround RUSTFLAGS=-Clinker=rust-lld; 29 tests total: apex-policy 5, apex-task-state 11, apex-evidence 13.
cargo clippy --manifest-path apexcode/Cargo.toml --workspace --all-targets -- -D warnings — PASS with the same linker workaround.
git diff --check — PASS.

NO CODEX RUNTIME INTEGRATION YET
NO PRODUCTION CHANGE
NO MERGE

Verification command correction

For the isolated virtual workspace, the exact formatting command is cargo fmt --manifest-path apexcode/Cargo.toml --all -- --check. The formatting-only closure commit makes this foundation PR fmt-clean.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit:9ae17e784d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-08-22T06:46:57Z

@@ -0,0 +1,7 @@
+[workspace]


Exercise the Apex workspace in CI
Adding this separate workspace does not make any existing PR job build or test it: .github/workflows/rust-ci.yml:43-47 only marks codex-rs/*, tooling, or workflow changes as relevant, and its Rust jobs run from codex-rs at lines 64-67. Consequently an Apex-only follow-up can merge with compilation, formatting, and all 29 tests skipped; add an Apex path detector and a job that checks this workspace.
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:57Z

@@ -0,0 +1,7 @@
+[workspace]


Ignore the Apex Cargo target directory
Running Cargo in the newly added workspace creates an untracked apexcode/target/ tree. A repository-wide search shows only codex-rs/.gitignore entries for that workspace's own /target/, so the documented Apex build workflow dirties the worktree and risks generated artifacts being committed; add an ignore rule scoped to apexcode/target/.
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:57Z

+#[cfg(test)]
+mod tests {


Move evidence tests into a sibling module
This is a newly introduced inline test module, whereas the repository requires new test modules to live in a descriptive sibling *_tests.rs file and be attached with an explicit #[path = "..."]; move these tests out of the implementation module.
AGENTS.md reference: AGENTS.md:L169-L178
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:57Z

+#[cfg(test)]
+mod tests {


Move task-state tests into a sibling module
This newly introduced inline test module should be placed in a descriptive sibling *_tests.rs file and referenced through an explicit #[path = "..."], as required for all new test modules in this repository.
AGENTS.md reference: AGENTS.md:L169-L178
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:57Z

+members = [
+ "apex-policy",
+ "apex-task-state",
+ "apex-evidence",


Split the independent foundation crates into stages
This non-mechanical commit changes 1,225 lines and combines three dependency-free crates with identity, security, fixture, and benchmark documentation, exceeding the repository's 800-line review limit. Land the smallest coherent stage first—such as the workspace plus apex-policy—then add task state, evidence, and documentation separately so their independent contracts remain reviewable.
AGENTS.md reference: AGENTS.md:L125-L131
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:57Z

+ let constraint_count = reader.u32().ok_or(SnapshotError::InvalidFormat)? as usize;
+ if constraint_count > reader.remaining() {
+ return Err(SnapshotError::InvalidFormat);
+ }
+ let mut constraints = Vec::with_capacity(constraint_count);


Bound constraint allocation before reserving
For malformed or corrupted snapshots, comparing constraint_count directly with the remaining byte count is insufficient because every constraint requires at least a four-byte length prefix and the snapshot also needs its status and revision. A roughly 100 MB input claiming 100 million constraints passes this check and Vec::with_capacity attempts about 2.4 GB of allocation before parsing fails, allowing recovery data to terminate the process through memory exhaustion; enforce the minimum encoded size and a hard count cap before reserving.
Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-08-22T06:46:58Z

+ let revision = self.revision.checked_add(1).ok_or(TransitionError {
+ from: self.status,
+ to: next,
+ })?;


Report revision exhaustion instead of an invalid transition
A structurally valid snapshot may decode with revision == u64::MAX, after which an otherwise allowed transition such as Planned -> Running fails here with TransitionError { Planned, Running } and is reported as an invalid state-machine edge. The dedicated TaskStateError::RevisionOverflow variant is never surfaced, so recovery code cannot distinguish revision exhaustion from an actually forbidden transition and the resumed task remains permanently stuck; return a distinct overflow error while preserving the state.
Useful? React with 👍 / 👎.

kratosboom added 10 commits August 22, 2026 12:44

feat(apex-policy): scaffold isolated policy crate
f83f5a1

feat(apex-policy): add initial risk and decision model
cfe1efe

docs(apexbench): define reproducible evaluation framework
6f015b5

docs(apex): identify ApexCode fork and bootstrap scope
6b5a6fd

docs(apex): extend security policy for ApexCode
1debda3

refactor(apex-policy): keep bootstrap crate dependency-free
f407f65

refactor(apex-policy): remove external serialization dependency
1732907

feat(apex): add persistent task state foundation
fc83a89

feat(apex): add revision-bound evidence gates
88924b5

docs(apex): document task and evidence contract
d3129b5

kratosboom mentioned this pull request Aug 22, 2026
feat(apex): add role-separated orchestration foundation #2
Draft

style(apex): normalize bootstrap rustfmt
9ae17e7

kratosboom marked this pull request as ready for review August 22, 2026 06:37

chatgpt-codex-connectorBot reviewed Aug 22, 2026
View reviewed changes

kratosboom marked this pull request as draft August 22, 2026 07:09

kratosboom added 2 commits August 22, 2026 14:15

fix(apex-evidence): fail closed on mixed evidence states
7d7890f

fix(apex-task-state): seal authoritative task mutation
07b14a9

kratosboom mentioned this pull request Aug 22, 2026
feat(apex): finalize ApexCode safety kernel and runtime gate v0.1 #6
Merged

kratosboom merged commit 67fdb80 into mainAug 22, 2026
2 checks passed

Navigation Menu

Uh oh!

feat(apex): bootstrap ApexCode safety and evaluation foundation - #1

feat(apex): bootstrap ApexCode safety and evaluation foundation#1
kratosboom merged 13 commits into
mainfrom
apex/bootstrap-v2

kratosboom commented Aug 22, 2026•
edited
Loading

Uh oh!

chatgpt-codex-connectorBot left a comment

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		#[cfg(test)]
		mod tests {

		#[cfg(test)]
		mod tests {

Uh oh!

Conversation

kratosboom commented Aug 22, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

What changes

Already established on main

Safety

Validation

Review gate

ApexCode foundation update

Verification command correction

Uh oh!

chatgpt-codex-connectorBot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connectorBotAug 22, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

kratosboom commented Aug 22, 2026•
edited
Loading