[senamakel]

Summary

A big hive.json is hard to read and edit. A hive may now move some of its manifest's top-level sections into three optional files at the package root, each a JSON object holding those sections under their usual keys:

File	May hold
agents.json	roles, seats, profiles
tools.json	tools, toolpacks, mcp, skills, hooks
desks.json	desks, topology, referrals

Parsing assembles them into the one Manifest before decoding (document/split/). Validation, lowering, patches, lineage and content identity therefore can't tell a split package from a single-file one. Moving a section between files is no change for check.

• Locations. Each split file is first decoded on its own as a typed struct that denies unknown fields, so a schema error, or a section the file doesn't own, is reported in that file with serde's line and column. After that, every finding about a moved section, from parsing, validation or check, names the split file at the same JSON Pointer (agents.json:/seats/0/role).
• One home per section. A section declared in both hive.json and its split file is a json finding at /<section> in the split file.
• check treats the split files as package files, not as files outside the package.

Public API / behaviour changes

• New optional package files agents.json, tools.json and desks.json. A package without them parses exactly as before.
• The outside_package message lists the new files.
• A schema error in hive.json's own sections of a split package has serde's message without a line and column, because the assembled manifest has no single source text. Single-file packages keep line and column.
• diagnose(&Package), which sees no files, still reports in hive.json. The serialized Package holds one manifest, so PackageSource::new writes one hive.json as before.

Validation

• cargo fmt --all -- --check: clean
• cargo clippy -p tinyhivemind-lang --all-targets -- -D warnings: clean
• cargo test -p tinyhivemind-lang: all pass, including the new tests/split_manifest.rs (7 tests: the gallery self-improving package split parses and lowers equal to the original, partial splits, duplicate sections, foreign sections and non-object files, findings located in the split file, check locating a forbidden edit in agents.json and giving the same review as the single-file edit, moving sections being no change) and document/split/test.rs (the typed files hold exactly their sections, each section has one owner, home).
• RUSTDOCFLAGS="-D warnings" cargo doc -p tinyhivemind-lang --no-deps: clean

Docs: docs/specs/hive-language.md gets a new "Splitting the manifest" section, and the crate overview table and document/README.md are updated. There is a new document/split/README.md.

[tinysweeper]

Tiny Sweeper review

Reviewed head: 2a901baa5439325f8608db6b3a366c6bec472299

Verdict: Ready for maintainer review · Priority: none
Coverage: Reviewed 11 of 11 files.
Pull request purpose (title): Let a hive split its…

Diagnostics and supporting details

Full summary

Pull request purpose (title): Let a hive split its manifest into agents.json, tools.json and desks.json. This pull request lets a hive split its manifest across three optional package-root files (`agents.json`, `tools.json`, `desks.json`) so a large `hive.json` is easier to read and edit. Parsing decodes each present split file on its own as a typed struct that denies unknown fields, then assembles the sections into one `Manifest` before decoding, so validation, lowering, patches and content identity cannot distinguish a split package from a single-file one. Findings about moved sections are relocated to the split file holding the section at the same JSON Pointer, and `check` treats the new files as package files.

Review snapshot

Change surface	Files	Review signal	Count
Production	6	Active findings	0
Tests	2	Noted findings	0
Documentation	3	Resolved findings	0
Configuration	0	Pending checks/questions	0

Completeness: Complete
Test assessment: Test coverage is assessed from changed tests and lane evidence; execution is not claimed without trusted check data.

What changed

Adds a `document/split` module defining the three split files, the sections each owns, their typed wire structs, assembly into the manifest, and a `home` function naming the split file a finding belongs in. `document::collect` delegates manifest decoding to `split::manifest` and relocates findings via the new `Findings::relocate`. `decode` is refactored around a new `decode_value` that tolerates a missing single source text, so an assembled manifest reports serde's message without line and column. In `check`, diagnostics are relocated to the split file the edit kept the section in, and `outside_package` no longer flags split files as unowned, with an updated message. Docs and shell-level documentation (`document/README.md`, `document/split/README.md`, `lib.rs` crate overview, and `docs/specs/hive-language.md`) describe the split and its location rules.

Features

• Added — Optional manifest split over agents.json, tools.json and desks.json: A hive may move roles/seats/profiles, tools/toolpacks/mcp/skills/hooks, or desks/topology/referrals out of `hive.json` into optional files at the package root, which are assembled into one manifest before decoding so downstream behavior is identical. (crates/tinyhivemind-lang/src/document/split/mod.rs, crates/tinyhivemind-lang/src/document/mod.rs pub(crate) fn collect(documents: &BTreeMap<String, String>) -> (Option<Package>,, docs/specs/hive-language.md and unsupported schema/version values fail rather than disappearing during a)
• Added — Per-file typed decoding with line and column errors: Each split file is decoded on its own as a struct that denies unknown fields, so a schema error or a section the file does not own is reported in that file with serde's line and column; an assembled manifest reports serde's message alone because it has no single source. (crates/tinyhivemind-lang/src/document/split/mod.rs, crates/tinyhivemind-lang/src/document/mod.rs fn decode<T: serde::de::DeserializeOwned + serde::Serialize>()
• Added — Findings relocated to the split file that holds the section: Findings from parsing, validation and `check` about a moved section name the split file at the same JSON Pointer (for example `agents.json:/seats/0/role`), so diagnostics point where the editor wrote the section. (crates/tinyhivemind-lang/src/diagnostic/mod.rs impl Findings {, crates/tinyhivemind-lang/src/check/mod.rs pub fn check(, crates/tinyhivemind-lang/src/document/split/mod.rs)
• Modified — check treats split files as package files: `outside_package` no longer reports changes to the new split files as unowned, and its message lists them among the files that are read. (crates/tinyhivemind-lang/src/check/rules.rs pub(super) fn outside_package(, crates/tinyhivemind-lang/src/document/README.md package's files into a `BTreeMap<String, String>` and calls `parse`; this)

Tests

• integration — The gallery `self-improving` package, split across the three files, parses and lowers equal to the one-file original, and its split `hive.json` no longer holds `seats` or `tools`.: The tests lane reports this real behavioural test asserts the split package parses and lowers identically to the single-file one and found no defect; execution was not verified by trusted check evidence. (crates/tinyhivemind-lang/tests/split_manifest.rs)

Findings

No active actionable findings.

Before merge

None.

Agent review details

critique

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The critique lane answered, reviewing 11 of 11 files with no findings listed.
• Lane summary: Reviewed 11 files; 2 findings. _The code index is behind this pull request (indexed at `f40d1c6488ee`), so retrieved context may be out of date._

security

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The security lane answered with no findings across 8 of 8 reviewable files; the three prose/spec files were not security-reviewed as prose or tabular data.
• Lane summary: Reviewed 8 files; 0 findings. 3 files were not security-reviewed: crates/tinyhivemind-lang/src/document/README.md (prose or tabular data), crates/tinyhivemind-lang/src/document/split/README.md (prose or tabular data), docs/specs/hive-language.md (prose or tabular data). _The code index is behind this pull request (indexed at `f40d1c6488ee`), so retrieved context may be out of date._

tests

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Lane summary: This adds optional manifest splitting over `agents.json`, `tools.json` and `desks.json`, relocating findings into the split file that holds the section, and is backed by a real behavioural test (`a_split_manifest_parses_and_lowers_like_one_hive_json`) asserting the split package parses and lowers identically to the single-file one, alongside tests for duplicate sections, unowned sections, relocated pointers, and check reporting. I found no defect in the diff: no regressions, no coverage gaps on the new branches, and no repository-rule violations. _The code index is behind this pull request (indexed at `f40d1c6488ee`), so retrieved context may be out of date._

commits

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The commits lane answered and reports no sensitive content in committed files.
• Lane summary: No sensitive content in committed files. _The code index is behind this pull request (indexed at `f40d1c6488ee`), so retrieved context may be out of date._

description

• Conclusion: Success
• Scope reviewed: all assigned evidence
• Positive: The description lane answered that the split is implemented as described, including assembly into one `Manifest`, relocation via `split::home` in both `diagnose_documents` and `check`, duplicate-section reporting in the split file, and the updated `outside_package` message, with nothing contradicting the stated contracts.
• Lane summary: The manifest split is implemented as described: optional `agents.json`/`tools.json`/`desks.json` (constants `AGENTS`/`TOOLS`/`DESKS`, table `SPLITS`) are decoded on their own, assembled into one `Manifest` before decoding, findings are relocated via `split::home` in both `diagnose_documents` and `check`, duplicate sections are reported in the split file, and `check` treats the new files as package files with an updated `outside_package` message. The module layout, docs and tests match the stated API changes, and I found nothing in the diff that contradicts the description or breaks a stated contract. _The code index is behind this pull request (indexed at `f40d1c6488ee`), so retrieved context may be out of date._

Evidence and run details

• Models: ladder/vectors-oai3, deepseek-v4.1-flash
• Spend: $0.001654
• Tokens: 1088170 input · 15382 output · 47616 cached · 766 embedding
• Updated: 2026-10-25T09:25:57Z

Head	State	Pass summary
2a901baa5439	ready for maintainer review	0 active finding(s), 0 resolved finding(s) (at 2026-10-25T09:25:57Z)

[coderabbitai]

Caution

Review failed

The pull request is closed.

Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	🔄 Running since 2026-10-25T09:24:40.218046Z	2a901ba	PR opened

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[tinysweeper]

tinysweeper found nothing blocking. Approving.

Diagnostics

Total spend: $0.0017