Reported issues for context-foundry

Pod holds 17 of 19 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to context-foundry.

Most discussed

MCP server not accessible globally from other directories

Problem

When running Claude Code from a different directory (e.g., from within bomb-man-adventure), the Context Foundry MCP server is not available. This prevents Claude Code from detecting intent keywords like "fix" and routing to the appropriate MCP functions.

The MCP server only works when Claude Code is running from within the context-foundry directory itself.

Root Cause

The MCP server configuration needs to be in the global Claude Code settings () with an absolute path, not just a l

Read the thread · 2025-10-26 · closed · 2 comments

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Metadata:

Type: Enhancement
Priority: P0
Category: testing, quality
Effort: Medium (1-3 days)
File: tools/mcp_server.py

Problem

No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

AI Analysis

Priority Score: 9/10

Why this matters: MCP server is the primary external interface for Context Foundry. Lack of tests creates security vulnerabilities, operational ri

Read the thread · 2025-11-11 · closed · 1 comment

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Metadata:

Type: Enhancement
Priority: P1
Category: testing, quality
Effort: Medium (1-3 days)
File: tools/mcp_server.py

Problem

No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

AI Analysis

Priority Score: 9/10

Why this matters: MCP server is a critical external interface handling protocol communication. Lack of tests creates security and reliability risk

Read the thread · 2025-11-11 · closed · 1 comment

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Metadata:

Type: Enhancement
Priority: P0
Category: testing, quality
Effort: Medium (1-3 days)
File: tools/mcp_server.py

Problem

No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

AI Analysis

Priority Score: 9/10

Why this matters: MCP server is the primary integration point for the entire system. Lack of tests for this critical component poses significant r

Read the thread · 2025-11-10 · closed · 1 comment

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Metadata:

Type: Enhancement
Priority: P0
Category: testing, quality
Effort: Medium (1-3 days)
File: tools/mcp_server.py

Problem

No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

AI Analysis

Priority Score: 9/10

Why this matters: MCP server is a core external-facing component that handles requests and integrations. Lack of tests creates significant reliabi

Read the thread · 2025-11-10 · closed · 1 comment

Security: Unsafe pickle usage - arbitrary code execution risk in core.py

🔒 Security: Unsafe pickle usage - arbitrary code execution risk in core.py

Metadata:

Type: Security
Priority: P0
Category: security, vulnerability
Effort: Small (< 1 day)
File: .venv-test/lib/python3.13/site-packages/diskcache/core.py (line 177)

Problem

Unsafe pickle usage - arbitrary code execution risk

Found at line 177 in .venv-test/lib/python3.13/site-packages/diskcache/core.py

Implementation Plan

Review and understand the issue
[

Read the thread · 2025-11-10 · closed · 1 comment

WIP(P32.8): End-to-end smoke test that exercises a real opencode -> LM S -- validation failed

WIP Commit: `P32.8`

Task: End-to-end smoke test that exercises a real opencode -> LM Studio round trip. Add scripts/smoke-local-model.sh (and a corresponding cargo test --test local_model_smoke) that spins up a tiny project, configures .foundry.json to route through opencode + lmstudio, invokes one stage end-to-end, and asserts the round trip worked. This is the GATE task -- no overnight session ships P32.x without it passing.

Validation did not pass. This issue was auto-created b

Read the thread · 2026-04-25 · open · 0 comments

WIP(P30.1): Document the recently-shipped task completion hook and pipel -- validation failed

WIP Commit: `P30.1`

Task: Document the recently-shipped task completion hook and pipeline_stages configuration in the project README. P28.1 added on_task_complete (a fire-and-forget shell hook with FOUNDRY_TASK_ID / FOUNDRY_TASK_STATUS / FOUNDRY_TASK_DESC / FOUNDRY_COMMIT_SHA env vars), but README.md does not mention it -- users must read src/config.rs:393-401 to discover the feature. Similarly P24.1 made pipeline_stages data-driven for display and enable/disable, but the README and docs

Read the thread · 2026-04-25 · open · 0 comments

Most recent

Add provenance tracking to reviewer findings

CCA Alignment: Task Statement 5.6

Current state: Reviewer findings include file, line, and issue description but don't carry source provenance -- which analysis step produced the finding, what evidence was consulted, or how the conclusion was reached.

Desired state: Attach source evidence metadata to each finding: the specific code that was analyzed, the reasoning chain, and cross-references to related findings. This gives downstream consumers (fixer, issue creator) full context.

Read the thread · 2026-03-19 · open · 0 comments

Expose pattern catalog and extensions as MCP resources

CCA Alignment: Task Statement 2.4

Current state: Patterns and extensions are loaded programmatically by the build loop. Agents have no way to browse the available knowledge catalog without the orchestrator injecting it.

Desired state: Expose the pattern catalog (titles, keywords, severity) and extension index as MCP resources. This gives agents visibility into what knowledge is available without requiring exploratory tool calls.

Why: Anthropic's CCA exam guide describes MCP res

Read the thread · 2026-03-19 · open · 0 comments

Reviewer agent reads .buildloop/logs/ files, hitting token limits

Problem

The reviewer agent explores the .buildloop/ directory and tries to read its own log files (.buildloop/logs/REVIEWER-*.jsonl, .buildloop/logs/FIXER-*.jsonl). These files can be 35-45k tokens, exceeding the 25k token read limit, causing errors and wasting context window on irrelevant data.

Observed Behavior

[tool] Read -- .buildloop/logs/REVIEWER-20260314-045507.jsonl
[stderr] Error: File content (35918 tokens) exceeds maximum allowed tokens (25000)
[tool] Read -- .build

[Read the thread](https://github.com/context-foundry/context-foundry/issues/198) · 2026-03-14 · closed · 0 comments

### Smart Build Mode: Pre-build consultation & dynamic phase selection

## Problem

Context Foundry currently treats every build request as "one size fits all" - spinning up all agents (Scout, Architect, Builder, Test, Screenshot, Documentation, Deploy, Feedback) regardless of project complexity. This leads to:

1. **Over-engineering simple projects** - A 3-file weather app doesn't need a full backend/frontend split, npm dev server, playwright testing, etc.
2. **Wasted tokens & time** - Running 8 phases when you only need 3
3. **No user control** - Can't easily say 

[Read the thread](https://github.com/context-foundry/context-foundry/issues/191) · 2025-12-20 · open · 0 comments

### Daemon fails when Claude CLI not in PATH

## Problem

When the Context Foundry daemon runs autonomous builds, it fails immediately if the `claude` CLI is not available in the daemon's PATH. This happens during the Scout phase when trying to execute subprocess commands.

## Error

[Errno 2] No such file or directory: 'claude'


## Root Cause
The phase execution system (`tools/mcp_utils/phase_execution.py`) and delegation system (`tools/mcp_utils/delegation.py`) directly call `subprocess.Popen(['claude', ...])` or `subprocess.run(['

[Read the thread](https://github.com/context-foundry/context-foundry/issues/178) · 2025-11-24 · closed · 0 comments

### Add tests for tools/mcp_server.py

## ✨ Add tests for tools/mcp_server.py

**Metadata:**
- **Type:** Enhancement
- **Priority:** P0
- **Category:** testing, quality
- **Effort:** Medium (1-3 days)
- **File:** `tools/mcp_server.py`

### Problem
No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

### AI Analysis
**Priority Score:** 9/10

**Why this matters:** MCP server is a critical external-facing component that handles inter-process communication. Lack of tests poses security, relia

[Read the thread](https://github.com/context-foundry/context-foundry/issues/151) · 2025-11-10 · closed · 0 comments

### Security: Shell injection risk with shell=True in test_scout_agent.py

## 🔒 Security: Shell injection risk with shell=True in test_scout_agent.py

**Metadata:**
- **Type:** Security
- **Priority:** P0
- **Category:** security, vulnerability
- **Effort:** Small (< 1 day)
- **File:** `tests/evolution/test_scout_agent.py` (line 275)

### Problem
Shell injection risk with shell=True

Found at line 275 in tests/evolution/test_scout_agent.py

### AI Analysis
**Priority Score:** 9/10

**Why this matters:** Shell injection in test code is critical - if tests run with elev

[Read the thread](https://github.com/context-foundry/context-foundry/issues/134) · 2025-11-10 · closed · 0 comments

### CRITICAL: Pattern library bootstrap failure - new users lose all community learnings

## Summary

The MCP server pattern storage system has a critical architectural flaw that prevents new users from benefiting from community-contributed patterns. When users clone Context Foundry, the project contains curated patterns in `.context-foundry/patterns/`, but the MCP server creates a new file at `~/.context-foundry/patterns/` and never merges the project patterns into global storage.

**Impact:** New users encounter all the same issues the community has already solved.

## Evidence

##

[Read the thread](https://github.com/context-foundry/context-foundry/issues/99) · 2025-11-09 · closed · 0 comments

### Add tests for tools/mcp_server.py

## ✨ Add tests for tools/mcp_server.py

**Metadata:**
- **Type:** Enhancement
- **Priority:** P2
- **Category:** testing, quality
- **Effort:** Medium (1-3 days)
- **File:** `tools/mcp_server.py`

### Problem
No test coverage found for mcp_server.py. File contains testable functions but lacks unit tests.

### Implementation Plan
- [ ] Review and understand the issue
- [ ] Implement fix/enhancement
- [ ] Add tests
- [ ] Update documentation if needed
- [ ] Create PR for review

---

🤖 *Generated 

[Read the thread](https://github.com/context-foundry/context-foundry/issues/57) · 2025-11-08 · closed · 0 comments

The remaining reports are on [the project's issue tracker](https://github.com/context-foundry/context-foundry/issues).

Reported issues for context-foundry

Most discussed

MCP server not accessible globally from other directories

Problem

Root Cause

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Problem

AI Analysis

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Problem

AI Analysis

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Problem

AI Analysis

Add tests for tools/mcp_server.py

✨ Add tests for tools/mcp_server.py

Problem

AI Analysis

Security: Unsafe pickle usage - arbitrary code execution risk in core.py

🔒 Security: Unsafe pickle usage - arbitrary code execution risk in core.py

Problem

Implementation Plan

WIP(P32.8): End-to-end smoke test that exercises a real opencode -> LM S -- validation failed

WIP Commit: P32.8

WIP(P30.1): Document the recently-shipped task completion hook and pipel -- validation failed

WIP Commit: P30.1

Most recent

Add provenance tracking to reviewer findings

CCA Alignment: Task Statement 5.6

Expose pattern catalog and extensions as MCP resources

CCA Alignment: Task Statement 2.4

Reviewer agent reads .buildloop/logs/ files, hitting token limits

Problem

Observed Behavior

WIP Commit: `P32.8`

WIP Commit: `P30.1`