[cli-tools-test] CLI bridge type coercion bugs: integer/array params fail; audit isError flag incorrect

[github-actions]

Daily exploratory testing session (§24597696067) identified several reproducible bugs in the audit, logs, and compile MCP tools. Testing was performed on 2026-04-18 against 194 workflows.

---

Bug 1 — logs --count N always fails (CLI bridge passes integers as strings)

Severity: High — count-based pagination is completely broken

The --count parameter is passed as a string "3" by the CLI bridge but the MCP tool's JSON Schema requires an integer. Every call with --count produces a validation error.

Reproduction:

agenticworkflows logs --count 3

Actual output:

validating "arguments": validating root: validating /properties/count:
  type: 3 has type "string", want "integer"

Expected: Logs limited to 3 results.

Impact: Users cannot paginate or limit log results. The CLI always downloads the full dataset regardless of what count value is specified.

---

Bug 2 — compile --workflows "name" always fails (CLI bridge passes arrays as strings)

Severity: High — targeted single-workflow compilation is completely broken

The --workflows parameter expects a JSON array but the CLI bridge passes the value as a plain string. There is no way to compile a specific workflow without compiling all 194.

Reproduction:

agenticworkflows compile --workflows "daily-issues-report"

Actual output:

validating "arguments": validating root: validating /properties/workflows:
  type: daily-issues-report has type "string", want one of "null, array"

Expected: Only daily-issues-report.md is compiled.

Impact: Every compile invocation must process all workflows, making targeted validation slow and expensive. Also blocks the max_tokens efficiency pattern recommended in workflow tooling documentation.

---

Bug 3 — audit error responses have isError: false instead of isError: true

Severity: Medium — breaks automated error detection

When audit is called with an invalid or non-existent run ID, it returns error content but sets isError: false in the MCP response envelope. Per the MCP specification, tool errors must set isError: true.

Reproduction:

agenticworkflows audit --run_id_or_url "9999999999"

Actual MCP response:

{
  "result": {
    "isError": false,
    "content": [{"type": "text", "text": "{\"error\":\"failed to audit workflow run: ✗ failed to fetch run metadata\",\"run_id_or_url\":\"9999999999\"}"}]
  }
}

Expected: "isError": true

Impact: Any client checking isError to distinguish success from failure will silently treat error responses as successes.

---

Bug 4 — logs cache directory not accessible to callers

Severity: Medium — logs output is unreadable

logs writes results to /tmp/gh-aw/logs-cache/<hash>.json and returns that path to the caller, but the directory has permissions that prevent access:

ls: cannot open directory '/tmp/gh-aw/logs-cache/': Permission denied

Unlike /tmp/gh-aw/aw-mcp/logs/ (which is accessible and pre-populated with run directories), the logs-cache/ directory used for inline tool responses is not readable. The logs tool's returned file_path field is therefore unusable.

Workaround: Use the pre-downloaded logs in /tmp/gh-aw/aw-mcp/logs/ directly.

---

What Works Correctly

View passing test results

• ✅ status — lists all 194 workflows with engine, trigger, and compile status
• ✅ audit — rich output for valid runs: overview, metrics, tool_usage, safe_output_summary, created_items, key_findings, recommendations, behavior_fingerprint
• ✅ audit safe outputs — correctly reports safe_output_summary (type breakdown) and created_items (with URLs and timestamps) for runs with safe output activity
• ✅ audit failed run — correctly identifies failure, reports high token usage and excessive turn count
• ✅ logs --workflow_name — filters to a named workflow (empty summary returned gracefully for non-existent workflows)
• ✅ logs --start_date / --end_date — date range filtering works; future dates and old dates handled gracefully (no crash)
• ✅ compile (no args) — compiles all 194 workflows, 0 errors, 0 warnings
• ✅ compile error detection — correctly reports invalid tool names and invalid YAML with line numbers and context
• ✅ compile YAML parse errors — reports exact line/column with surrounding context

---

Environment

• Repository: github/gh-aw
• Run ID: §24597696067
• Date: 2026-04-18
• Workflows tested: 194
• Audit runs tested: 4 (success, failure, safe-outputs, invalid ID)

Generated by Daily CLI Tools Exploratory Tester · ● 3.1M · ◷

• expires on Apr 25, 2026, 5:32 AM UTC