chore(deps): integrate amplihack-rs v0.18.25

Cargo.toml

@@ -154,11 +154,53 @@ rustyclawd-tools = { git = "https://github.com/rysweet/RustyClawd.git", rev = "d
 # wiring (`src/gym_runner_client.rs`) and Simard-specific scoring/history stay
 # in this crate. The crate is a light workspace member (serde/serde_json/
 # thiserror/tracing/chrono only — no heavy workspace deps are transitive).
-# Bumped (issue #2626) to the current HEAD of amplihack-rs `main` (verified via
-# `git ls-remote`); the delta is workflow/recipe/CI-tooling fixes only, with the
-# `amplihack-agent-eval` crate surface (GymRunner/GymConfig/GymScenarioResult)
-# and the reverted workspace version 0.11.1 unchanged.
-amplihack-agent-eval = { git = "https://github.com/rysweet/amplihack-rs.git", rev = "14dc30b10e87764120c6f2bae7f3630522c29e5d" }
+# Pin lineage (each rev is an immutable amplihack-rs commit SHA):
+#   59548a96… (initial de-fork, #2323)
+#     -> 2a93441d… landed by issue #2626 — bump both amplihack pins to the
+#        then-current upstream `main` HEAD;
+#     -> 14dc30b1… landed by issue #2767 — pick up amplihack-rs PR #856, the
+#        clean agent-result channel off the noisy stdout (a 2026-07-07 UTC
+#        commit: 2026-07-07 01:47:01 UTC / 2026-07-06 18:47:01 -0700);
+#     -> 9ee05a06… (this pin, below).
+# Across 2a93441d…/14dc30b1… the delta was workflow/recipe/CI-tooling only, with
+# the `amplihack-agent-eval` crate surface (GymRunner/GymConfig/
+# GymScenarioResult) unchanged.
+#
+# This pin is the amplihack-rs **v0.18.25** release source commit
+# `9ee05a06eab98e9ab504a031bffaa4190700c2af` (release published 2026-09-02;
+# pin verified 2026-09-03). Provenance, all verified with `git ls-remote` /
+# `git cat-file` against https://github.com/rysweet/amplihack-rs.git:
+#   * the **annotated** tag `refs/tags/v0.18.25` is object `e947170a…`, and it
+#     dereferences (`refs/tags/v0.18.25^{}`) to commit `9ee05a06…`;
+#   * `refs/heads/main` *was* `9ee05a06…` **at verification on 2026-09-03** —
+#     i.e. the release tag and `main` pointed at the same commit at that
+#     instant, which establishes that `9ee05a06…` is a mainline release commit
+#     rather than a side/topic branch. This is a point-in-time observation, not
+#     a standing claim: upstream `main` advances, and once it does this comment
+#     stays true (it records what was observed then) while the pin deliberately
+#     does not move.
+#   * `9ee05a06…` was 200 commits ahead of the previous pin `14dc30b1…` at that
+#     same verification.
+# The pin stays the immutable commit SHA (never the tag) so the build cannot be
+# re-pointed by an upstream tag move.
+#
+# API impact: NONE. The `crates/amplihack-agent-eval` subtree is **byte
+# identical** across those 200 commits — both `14dc30b1…:crates/amplihack-agent-eval`
+# and `9ee05a06…:crates/amplihack-agent-eval` resolve to git tree
+# `1a635333f15fa2964e6d0ecae35b5a9625d59ee7`, and the crate's own dependency set
+# (serde/serde_json/thiserror/tracing/chrono) is unchanged. So
+# `src/gym_runner_client.rs` needs no adaptation and the wire JSON stays
+# byte-stable.
+#
+# Lockfile impact: the crate inherits `version.workspace`, which upstream moved
+# 0.11.1 -> 0.18.0 in the workspace manifest, so Cargo.lock records
+# `amplihack-agent-eval 0.18.0` at the new rev. NOTE the deliberate mismatch:
+# the upstream *release* is tagged v0.18.25 while the upstream *workspace crate
+# version* at `9ee05a06…` is 0.18.0 — the tag tracks the release train, not the
+# crate version. Both readings are pinned to that immutable commit, so neither
+# changes as upstream advances. Simard consumes the crate by rev, so only the
+# rev is authoritative.
+amplihack-agent-eval = { git = "https://github.com/rysweet/amplihack-rs.git", rev = "9ee05a06eab98e9ab504a031bffaa4190700c2af" }
 # De-fork phase 2b (issue #2307): the native cognitive-memory fork is gone, but
 # `lbug` is still a direct dependency of the standalone `simard-tui` binary
 # (`src/bin/simard_tui/goals.rs`), which opens a LadybugDB read-only to render

docs/architecture/gym-eval-library-adapter.md

@@ -1,7 +1,7 @@
 ---
 title: Library-backed Gym Evaluation Engine (the sole engine)
 description: How Simard's gym client is backed by the amplihack-agent-eval crate's native Rust GymRunner through the thin gym_runner_client adapter. As of the de-fork the private native_gym reimplementation has been deleted and the library is the only evaluation engine.
-last_updated: 2026-06-20
+last_updated: 2026-09-03
 owner: simard
 doc_type: reference
 related:
@@ -64,11 +64,15 @@ three handlers moved into the library.
 ## The dependency
 
 The engine is a pinned-revision git dependency in `Cargo.toml`, matching the
-immutable-rev style already used for `amplihack-memory` and `rustyclawd-core`:
+immutable-rev style already used for `amplihack-memory` and `rustyclawd-core`.
+The authoritative pin is always the live line in the root `Cargo.toml`; the
+snippet below records the rev current as of **2026-09-03** — the amplihack-rs
+`v0.18.25` release source commit (the annotated tag `v0.18.25` dereferences to
+it, and it was `main` HEAD at verification):
 
 ```toml
 [dependencies]
-amplihack-agent-eval = { git = "https://github.com/rysweet/amplihack-rs.git", rev = "59548a96049ab8d558110bcaf9c82a4316f1bbf0" }
+amplihack-agent-eval = { git = "https://github.com/rysweet/amplihack-rs.git", rev = "9ee05a06eab98e9ab504a031bffaa4190700c2af" }
 ```
 
 ### Consumability
@@ -195,11 +199,19 @@ scenario list. The wire `scenario_id` is therefore always one of the ids
 `gym.list_scenarios` advertises — never the bare `"L{n}"` form.
 
 **2. Suite `success`.** The library's `run_suite` computes its top-level
-`success` with inverted logic —
+`success` as
 `!result.failed_levels.is_empty() || result.level_results.iter().all(|lr| lr.success)`
-(`amplihack_agent_eval::gym::GymRunner::run_suite`, gym.rs:300-301) — which
-evaluates to `true` precisely when there *are* failed levels. The adapter does
-**not** trust that flag; it recomputes
+(`amplihack_agent_eval::gym::GymRunner::run_suite`). This is a **tautology: it
+always evaluates to `true`.** `ProgressiveResult::add_result` records a level id
+in `failed_levels` exactly when that level's result is *not* successful, so an
+empty `failed_levels` implies every entry of `level_results` succeeded — meaning
+whenever the first disjunct is `false`, the second is `true`. The flag is
+therefore `true` on an all-pass suite, on a partially-failing suite, and on an
+empty suite (`all()` over an empty iterator is `true`).
+
+The consequence is stronger than "the flag is inverted": it carries **no
+information at all**, so there is nothing to invert or correct. The adapter
+ignores it outright and recomputes
 `success = scenarios_passed == scenarios_total` from the per-scenario results it
 already maps. (This upstream quirk should be filed as an issue against
 `amplihack-rs`; until it is fixed, the adapter's recomputation is the contract

docs/atlas/compile-deps/README.md

@@ -1,3 +1,13 @@
+---
+title: Compile Dependencies Atlas
+description: "Code-derived atlas layer mapping Simard's direct Cargo dependencies (including the exact git-rev pins) and a one-hop view of the largest internal Rust module dependencies."
+last_updated: 2026-09-03
+review_schedule: as-needed
+owner: simard
+doc_type: reference
+status: living document — regenerate from code when stale
+---
+
 # Compile Dependencies Atlas
 
 This layer maps Simard's direct Cargo dependencies and a one-hop view of the largest internal Rust module dependencies. The overview diagrams stay small by grouping the external crates, dev/build dependencies, and internal module graph; the split diagrams below expand those groups without exceeding the atlas density target.
@@ -22,20 +32,40 @@ This layer maps Simard's direct Cargo dependencies and a one-hop view of the lar
 
 ## Evidence anchors
 
-- Direct dependencies are declared in `Cargo.toml:135` through `Cargo.toml:209`; Unix target dependency is at `Cargo.toml:212`; build and dev dependencies are at `Cargo.toml:218` and `Cargo.toml:245` through `Cargo.toml:254`.
-- `amplihack-memory` is pinned to `amplihack-memory-lib` with the `persistent` feature at `Cargo.toml:135`; the direct `lbug` fork pin is at `Cargo.toml:158`.
-- The default feature set enables `signal` and `dashboard-audit` at `Cargo.toml:227`, and `dashboard-audit` pulls `headless_chrome`, `regex`, and `url` at `Cargo.toml:236`.
-- The large internal module declarations are visible in `src/lib.rs:24`, `src/lib.rs:60`, `src/lib.rs:78`, `src/lib.rs:89`, `src/lib.rs:90`, `src/lib.rs:103`, `src/lib.rs:106`, `src/lib.rs:114`, `src/lib.rs:117`, `src/lib.rs:122`, `src/lib.rs:123`, `src/lib.rs:124`, `src/lib.rs:126`, `src/lib.rs:128`, `src/lib.rs:138`, `src/lib.rs:163`, and `src/lib.rs:164`.
+Anchors are **section- and symbol-based**, not line numbers: `Cargo.toml` and
+`src/lib.rs` both carry long provenance comment blocks, so raw line numbers rot
+on every pin bump (they were already stale before this page was last touched).
+Each anchor below is a stable, greppable construct.
+
+- Direct dependencies are declared under the `[dependencies]` table in
+  `Cargo.toml`; the Unix-only dependency (`libc`) is under
+  `[target.'cfg(unix)'.dependencies]`; build and dev dependencies are under
+  `[build-dependencies]` (`chrono`) and `[dev-dependencies]` (`assert_cmd`,
+  `proptest`, `serial_test`, `syn`, `proc-macro2`).
+- `amplihack-memory` is pinned to `amplihack-memory-lib` with the `persistent`
+  feature on the `amplihack-memory = { git = …, features = ["persistent"] }`
+  key; the direct `lbug` fork pin is the `lbug = { git = …, rev = … }` key. Both
+  are in `[dependencies]` — grep the key names.
+- The `[features]` table sets `default = ["signal", "dashboard-audit"]`, and
+  `dashboard-audit = ["dep:headless_chrome", "dep:regex", "dep:url"]`.
+- The large internal modules the module-graph diagram draws are the `pub mod` /
+  `mod` declarations in `src/lib.rs` — e.g. `cognitive_memory`, `goal_curation`,
+  `ooda_loop`, `ooda_actions`, `ooda_brain`, `journal`, `overseer`,
+  `meeting_backend`, `memory_consolidation`, `memory_ipc`, `engineer_loop`,
+  `self_deploy`, `self_improve`, and `runtime`. Every node name in
+  `compile-deps-internal-modules.dot` matches a `mod <name>;` declaration in
+  `src/lib.rs`, so the diagram can be re-checked with
+  `grep -E '^(pub )?mod <name>;' src/lib.rs`.
 
 ## Dependency inventory
 
 | Crate or section | Version or pin | Purpose |
 | --- | --- | --- |
-| `amplihack-memory` | git `c266e15d1399967c04324370e77cf281990b8be1`, feature `persistent` | Sole cognitive-memory backend adapter; persistent graph storage via upstream memory library. |
+| `amplihack-memory` | git `0031505b911151bf47409694a6c45f8b778d91b9`, feature `persistent` | Sole cognitive-memory backend adapter; persistent graph storage via upstream memory library. |
 | `lbug` | git `5a2c107881879f4d1bb594b14967948870e65cdc` | Embedded LadybugDB graph store used directly by TUI and unified with `amplihack-memory`. |
 | `rustyclawd-core` | git `dcccad80ed381c66a7728565be5cb84120aacbed` | RustyClawd core agent SDK integration. |
 | `rustyclawd-tools` | git `dcccad80ed381c66a7728565be5cb84120aacbed` | RustyClawd tool integration. |
-| `amplihack-agent-eval` | git `14dc30b10e87764120c6f2bae7f3630522c29e5d` | Native Rust gym and evaluation runner types. |
+| `amplihack-agent-eval` | git `9ee05a06eab98e9ab504a031bffaa4190700c2af` (amplihack-rs release `v0.18.25`) | Native Rust gym and evaluation runner types. |
 | `rusqlite` | `=0.31.0`, features `backup`, `bundled` | SQLite storage and backup support. |
 | `serde` | `=1.0.228`, feature `derive` | Serialization derives for config, state, and message types. |
 | `serde_json` | `=1.0.149` | JSON IO for CLI, recipes, telemetry payloads, and stored records. |

docs/atlas/compile-deps/compile-deps-external-runtime.dot

@@ -5,10 +5,10 @@ digraph compile_deps_external_runtime {
 
     simard [label="simard"];
 
-    amplihack_memory [label="amplihack-memory\ngit c266e15 persistent"];
+    amplihack_memory [label="amplihack-memory\ngit 0031505 persistent"];
     rustyclawd_core [label="rustyclawd-core\ngit dcccad8"];
     rustyclawd_tools [label="rustyclawd-tools\ngit dcccad8"];
-    amplihack_agent_eval [label="amplihack-agent-eval\ngit 14dc30b"];
+    amplihack_agent_eval [label="amplihack-agent-eval\ngit 9ee05a0"];
     lbug [label="lbug\ngit 5a2c107"];
 
     rusqlite [label="rusqlite =0.31.0\nbackup bundled"];

docs/atlas/compile-deps/compile-deps-external-runtime.mmd

@@ -1,8 +1,8 @@
 flowchart LR
-    simard["simard"] --> amplihack_memory["amplihack-memory git c266e15 persistent"]
+    simard["simard"] --> amplihack_memory["amplihack-memory git 0031505 persistent"]
     simard --> rustyclawd_core["rustyclawd-core git dcccad8"]
     simard --> rustyclawd_tools["rustyclawd-tools git dcccad8"]
-    simard --> amplihack_agent_eval["amplihack-agent-eval git 14dc30b"]
+    simard --> amplihack_agent_eval["amplihack-agent-eval git 9ee05a0"]
     simard --> lbug["lbug git 5a2c107"]
     simard --> rusqlite["rusqlite =0.31.0 backup bundled"]
     simard --> serde["serde =1.0.228 derive"]