Fix describe_outputs crash when tool_state is a dict not a JSON string

[jmchilton]

Problem

describe_outputs (planemo/galaxy/workflows.py) read step tool_state with an unconditional json.loads:

tool_state = json.loads(step.get("tool_state", "{}"))

Some .ga workflows carry tool_state as an already-decoded dict rather than a JSON-encoded string. On those, json.loads raises:

TypeError: the JSON object must be str, bytes or bytearray, not dict

This fires on the planemo test / report path (via get_outputs → describe_outputs) for a workflow that otherwise runs green.

Fix

Guard on type before decoding — the exact pattern already merged for the lint path in #1626, which missed this second call site. This keeps workflow_lint and planemo test consistent in how they tolerate dict tool_state:

raw_tool_state = step.get("tool_state", {})
if isinstance(raw_tool_state, str):
    tool_state = json.loads(raw_tool_state)
else:
    tool_state = raw_tool_state

Test

Adds test_describe_outputs_dict_tool_state, reusing the existing wf14-unlinted-best-practices-dict-tool-state.ga fixture (its step 0 has tool_id: null + dict tool_state — exactly this path). Verified red→green: without the fix the test reproduces the TypeError; with it, the file's tests pass.

🤖 Generated with Claude Code