Reported issues for automatised-pipeline

Pod holds 16 of 16 GitHub reports that passed its relevance review. This can include external user reports, maintainer-confirmed bugs, and concrete feature gaps. Treat them as evidence to inspect, not a count of distinct defects.

Back to automatised-pipeline.

Most discussed

bug(distribution): live-mount dev symlink is structurally incompatible with the marketplace digest pin — server dies with 'cached binary digest mismatch'

Symptom

The plugin MCP server fails to connect in Claude Code (✘ Failed to connect — -32000: Connection closed). Manual launch with CLAUDE_PLUGIN_ROOT set reproduces:

ai-architect-mcp-codebase: FATAL: cached binary digest mismatch; reinstall the plugin

Observed 2026-08-07 on the 0.9.0 marketplace install. This is also the root cause of BOTH tool-availability caveats in the A/B bench rev.2 report (harness-comparison/results/COMPARISON-rev2.md): the cortex-viz probe session…

Read the thread · 2026-08-07 · closed · 1 comment

distribution: ship a SessionStart/SubagentStart code-discovery nudge — codebase-memory-mcp captures tool preference globally and ai-architect ships nothing

Measured on this machine 2026-08-06. codebase-memory-mcp (the competitor tracked in the CBM parity arc) has installed itself into the user's global ~/.claude/settings.json, not just its own plugin manifest:

PreToolUse:     matcher "Grep|Glob"  -> ~/.claude/hooks/cbm-code-discovery-gate
SessionStart:   matchers startup / resume / clear / compact -> ~/.claude/hooks/cbm-session-reminder
SubagentStart:  matcher "*"          -> ~/.claude/hooks/cbm-subagent-reminder

The SessionStart…

Read the thread · 2026-08-06 · open · 1 comment

analyze_codebase fails with 'unknown relationship type: Uses_Field_TypeAlias' on real Rust codebases

Summary

analyze_codebase fails on any Rust codebase that uses a type alias as a struct/enum field type. The resolver constructs an edge table name dynamically (Uses_Field_<TargetLabel>) but the schema only registers Uses_Field_Struct, Uses_Field_Enum, and Uses_Field_Trait — never Uses_Field_TypeAlias.

The pipeline's own src/ triggers it.

Reproduction

Drove the MCP binary directly via stdio JSON-RPC against the pipeline's own source:


[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/1) · 2026-04-17 · closed · 1 comment

### index_status reports zero counts, without an error, when the graph is held by another request

Found by the review of PR #360 (issue #352), by reading the code; not reproduced yet.

`index_status` reads its counts through `graph_counts`, which ends in `unwrap_or_default` (src/history_handlers.rs). Since #360, opening a graph while another request in the same process holds the cached handle is refused with `graph_handle_in_use` or `graph_cache_busy`. For the write tools that refusal is an error. For `index_status` the refusal is swallowed by `unwrap_or_default`, so the tool answers…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/361) · 2026-09-25 · closed · 0 comments

### resolver(rust): a return-type hint promoted after an external use stays accepted when a later incremental run changes the crate names

## Summary

When the return-type rule of #348 accepts an explicit `use` because its first segment is a crate of the repository, the result is stored in the call site's hint and never re-checked. A later incremental run that changes the crate names in `Cargo.toml` does not re-examine the files it did not touch, so their hints stay accepted under the old names. The graph then depends on the order of the runs: an incremental run and a fresh full run over the same tree disagree.

## Context

Found…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/358) · 2026-09-24 · closed · 0 comments

### resolver(rust): use crate::X inside a tests/, benches/, examples/ or bin target is trusted by name, so a re-exported external type gets an edge to a library homonym

## Summary

Inside a Cargo test, bench, example or binary target, `crate` names that target, not the library. The receiver-from-return-type rule of #348 accepts `use crate::Set;` as evidence that `Set` is a type of the repository, by name only. A test file that re-exports an external `Set` at its root and reaches it through `crate::Set` gets an edge to the library's unrelated `Set::answer` at confidence 0.85.

## Context

Found by reading the code during the review of PR #351 (which introduced…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/357) · 2026-09-24 · closed · 0 comments

### graph: calls to a tuple-struct constructor are marked resolved but have no per-site row and no edge, because no relationship table targets a Struct

## Summary

A call to a tuple-struct constructor, such as `Tier(1)`, is marked `is_resolved = true` but gets no row in any `Calls_CallSite_*` table and no function-level `Calls_*` edge. The schema has no relationship table whose target is a `Struct`, so the resolution has nowhere to go. #335 closed the same gap for calls that resolve to a function or a method and left this one open on purpose, because covering it needs a schema change.

## Context

Noted while fixing #335 (PR #337): calls to a…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/356) · 2026-09-24 · closed · 0 comments

### resolver(rust): a receiver that is a tuple-struct constructor, a struct literal or Type::new(..) written in place is left unresolved statically

## Summary

The static resolver types a receiver bound by `let t = Type::new(..)` but not one bound by `let t = Tier(1)` (a tuple-struct constructor) or `let v = Named { n: 3 }` (a struct literal), and not a receiver that is written in place: `Tier(1).join(..)`, `Tier::new(1).join(..)`, `Named { n: 3 }.get()`. The type is written in the expression in every one of these forms. `lsp_resolve` resolves all of them; a static-only run misses them.

## Context

Found by replaying failure shapes from…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/355) · 2026-09-24 · closed · 0 comments

## Most recent

### get_impact: test, bench and example callers are counted with production callers, and entry_kind marks only plain #[test]

## Summary

`get_impact` lists test functions, bench code and example code among the callers of a production function, in the same list, with the same fields, and counts them in `callers_total`. A reader cannot tell how many callers are production code. The graph does carry a marker for tests (`Function.entry_kind = 'test'`), but it is set only for a plain `#[test]` attribute, and `get_impact` does not use it.

## Context

Found by replaying failure shapes from the public tracker of…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/354) · 2026-09-24 · closed · 0 comments

### coverage: an unknown Cargo target map renders identically to a fully-compiled one

Follow-up from the 2026-09-22 AxonOS re-verification of #282/#284 (`tasks/axon-verify-20260922/REPORT.md`, finding F2).

## Measured

In the #282 nested-workspace shape, `cargo metadata --no-deps` fails with `current package believes it's in a workspace when it's not`. Per the #284 design (`tasks/plan-issues-282-283-284.md` section 3.2: if `cargo` is absent, or there is no `Cargo.toml`, or the command fails, the target map becomes `Unknown` and no file is attributed, an honest absence), no file…

[Read the thread](https://github.com/cdeust/ai-architect-mcp-codebase/issues/316) · 2026-09-22 · closed · 0 comments

### lsp: analysing a workspace root outside every compiled target still reports state=completed

Follow-up from the 2026-09-22 AxonOS re-verification of #282 (`tasks/axon-verify-20260922/REPORT.md`, finding F1).

## Measured

`analyze_codebase(path=<nested-nogit>/, lsp=true)` against a virtual `[workspace] members = []` root whose only child (`dy-wcet/`) is not listed as a member (`raw/lsp-parent-root-nogit/analyze.json`, commit `22b11da` on branch `docs/axon-verify-20260922`):

lsp_status = { state: "completed", server_health: {health: "warning", message: "Failed to read Cargo…

Read the thread · 2026-09-22 · closed · 0 comments

indexer: 270 of 818 tracked files absent from the index, none of them reported, and src/bin pruned as build output

Measured on this repository on 2026-09-09 with the release binary at 04e326a:

tracked files                 818
files_indexed (reported)      549
files with a recorded gap      10
absent from the manifest      270

270 tracked files were absent from the index and not one of them appeared anywhere in the coverage report. The run reported status: ok.

A. The walk prunes a Cargo source directory

bin is in the built-in prune list under "General build output" beside dist and…

Read the thread · 2026-09-09 · closed · 0 comments

Calls from files outside compiled Cargo targets are never resolved — Kani proof harnesses are structurally invisible to the call graph

Summary

Source files that are not part of a compiled Cargo target are parsed and indexed — their declarations are recovered correctly — but their outgoing calls are never resolved, even with rust-analyzer enabled. Verification harnesses are the canonical case: Kani proof files live outside src//tests/ and are compiled only under cargo kani.

The consequence is a graph that knows a proof harness exists but not that it exercises the function it proves. For a tool whose…

Read the thread · 2026-09-08 · closed · 0 comments

LSP resolution silently resolves nothing when the target sits under a parent Cargo workspace — reports state=completed, status=ok

Summary

When the analyzed codebase sits inside a parent Cargo workspace that does not list it as a member, the LSP resolution phase resolves nothing — and reports success anyway. lsp_status.state is "completed", the envelope status is "ok", and the coverage sidecar flags no gap. The consumer receives a graph whose audited projection is byte-identical to a static (no-LSP) run, with no signal that the phase it explicitly requested did nothing.

On Rust, where static resolution…

Read the thread · 2026-09-08 · closed · 0 comments

feat(index_codebase): user-configurable folder exclusion — an unreadable secrets directory aborts the entire walk

Problem

A user cannot complete index_codebase on their repository: the tree contains a directory that is locked down because it holds secrets (permission-denied at the filesystem level), and the indexer has no way to route around it.

Two distinct gaps compound into a hard failure:

No user-facing exclusion parameter. The index_codebase input schema (src/tool_schemas.rs, index_codebase_schema) exposes language, dependency_scope, bootstrap, full, cochange, … but…

Read the thread · 2026-08-13 · closed · 0 comments

SECURITY.md points at a private advisory channel that is disabled — no working private disclosure path

SECURITY.md tells reporters to open a private GitHub security advisory, but that channel is switched off, so a non-collaborator who follows the instructions has nowhere to go.

Evidence (measured 2026-07-27)

$ gh api repos/cdeust/automatised-pipeline/private-vulnerability-reporting
{"enabled":false}

SECURITY.md's only disclosure channel is https://github.com/cdeust/automatised-pipeline/security/advisories/new. With private vulnerability reporting disabled, that form is…

Read the thread · 2026-07-27 · closed · 0 comments

The remaining reports are on the project's issue tracker.