File: scripts/sw-loop.sh:561-608 — _extract_text_from_json() function

The execution path when Claude outputs a JSON object ({...}) instead of an array ([...]):

1. Line 579 — Case 2 checks first_char == "[" → false (it's {). The entire jq extraction block is skipped, even though jq is available.
2. Line 599 — Case 3 catches first_char == "{" and unconditionally prints "JSON output but jq not available". This is misleading because jq was never tried — the condition above only checked for arrays.

1. Most likely (confirmed): Case 2 only handles JSON arrays ([). When Claude's --output-format json outputs an object ({...}), the code falls through to Case 3 which prints a misleading "jq not available" warning regardless of whether jq is installed.
2. Less likely: Claude always outputs arrays and the { case never fires. Disproved — the issue reporter observed this in practice.
3. Unlikely: jq is actually unavailable. Disproved — the warning text itself is the problem, not jq availability.

• Pros: Actually parses the JSON object and extracts .result, providing the correct text output. Fixes both the misleading warning AND extracts useful content.
• Cons: Slightly more code, but the logic is straightforward.
• Blast radius: Minimal — only changes the _extract_text_from_json function. The fallback behavior (Case 3) still exists for the rare case where jq is truly unavailable.
• Complexity: Low — adds one elif branch with a parallel jq extraction for objects.

• Pros: Smallest possible change (1-2 lines).
• Cons: Doesn't actually parse the JSON object — still uses raw output even when jq could extract .result. Fixes the symptom but not the underlying limitation.
• Blast radius: Trivial.

Option A is the right fix because it's still a small change (10-15 lines) and it actually solves the problem properly — extracting .result from JSON objects the same way we extract from arrays. The warning fix in Case 3 comes along as part of the change.

1. scripts/sw-loop.sh — Fix _extract_text_from_json() to handle JSON objects
2. scripts/sw-loop-test.sh — Add test case for JSON object extraction

Restructure Cases 2 and 3 to:

• Case 2a: JSON array ([) with jq → extract .[-1].result (unchanged)
• Case 2b: JSON object ({) with jq → extract .result directly
• Case 3: JSON but no jq → accurate warning message

# Case 2: Valid JSON — extract .result with jq
if [[ "$first_char" == "[" ]] && command -v jq >/dev/null 2>&1; then
 # ... existing array extraction (unchanged) ...
elif [[ "$first_char" == "{" ]] && command -v jq >/dev/null 2>&1; then
 local extracted
 extracted=$(jq -r '.result // empty' "$json_file" 2>/dev/null) || true
 if [[ -n "$extracted" ]]; then
 echo "$extracted" > "$log_file"
 return 0
 fi
 # Try .content fallback for objects too
 extracted=$(jq -r '.content // empty' "$json_file" 2>/dev/null) || true
 if [[ -n "$extracted" ]]; then
 echo "$extracted" > "$log_file"
 return 0
 fi
 warn "JSON object output has no .result field — check $json_file"
 echo "(no text result in JSON output)" > "$log_file"
 return 0
fi
# Case 3: Looks like JSON but no jq — can't parse, use raw
if [[ "$first_char" == "[" || "$first_char" == "{" ]]; then
 warn "JSON output but jq not available — using raw output"
 cp "$json_file" "$log_file"
 return 0
fi

Key: Case 3's warning is now accurate — it only fires when jq is genuinely unavailable (both Case 2 branches check command -v jq first).

Add a new test in the "json extraction robustness" section (after Test 19, around line 299) that tests JSON object input:

# Test: JSON object → extracts .result
echo '{"type":"result","subtype":"success","result":"Object extraction works"}' > '$tmpdir/obj.json'
_extract_text_from_json '$tmpdir/obj.json' '$tmpdir/out_obj.log' ''

Then assert grep -q "Object extraction works" "$tmpdir/out_obj.log".

./scripts/sw-loop-test.sh

• Task 1: Read and understand the current _extract_text_from_json function (lines 561-608)
• Task 2: Add elif branch for JSON objects ({) with jq extraction in sw-loop.sh
• Task 3: Verify Case 3 warning now only fires when jq is genuinely unavailable
• Task 4: Add test case for JSON object extraction in sw-loop-test.sh
• Task 5: Run sw-loop-test.sh and verify all tests pass
• Task 6: Manually verify the fix handles edge cases (empty .result, missing .result)

Task dependencies: Task 1 → Task 2 → Task 3 (sequential). Task 4 depends on Task 2. Task 5 depends on Tasks 2-4. Task 6 depends on Task 5.

1. Existing tests — Run sw-loop-test.sh to ensure no regressions in array extraction, empty file handling, plain text pass-through, nested JSON, and binary input.
2. New test — JSON object with .result field → verify extraction works.
3. Edge cases covered by existing fallbacks:
   • JSON object without .result → warning + placeholder (same behavior as array without .result)
   • No jq available → accurate "jq not available" warning + raw output
   • Non-JSON input → pass-through (unchanged Case 4)

• _extract_text_from_json correctly extracts .result from JSON objects ({...})
• _extract_text_from_json still correctly extracts .result from JSON arrays ([...])
• The misleading "jq not available" warning no longer fires when jq IS available
• The "jq not available" warning still fires when jq is genuinely not installed
• New test case for JSON object extraction passes
• All existing sw-loop-test.sh tests pass
• No Bash 3.2 incompatibilities introduced

1. Run ./scripts/sw-loop-test.sh — all tests pass including the new JSON object test
2. Confirm the warning text change: grep for the old misleading message to ensure it's gone from the jq-available path
3. Confirm Case 3 only fires when command -v jq fails (structural code review)

Not applicable — this is a bash function fix, not an API change.

Not applicable — no API endpoints.

Not applicable.

Not applicable — internal function change, no public API.