fix(autonomous): guard rowTo* JSON.parse with safe fallback (AUDIT-H1)

[konard]

Fixes #260 — AUDIT-H1: JSON.parse in rowTo* has no try/catch — one bad row DoSes listing.

Problem

rowToTask, rowToCheckpoint, and rowToLogEntry in src/memory/agent/autonomous-tasks.ts called JSON.parse(...) on ten different columns without any error handling. A single row with corrupt JSON (manual DB edits, crashed writes, buggy backfills) propagated a SyntaxError out of listTasks / getTask / getCheckpoint / getExecutionLogs, breaking the entire /api/autonomous page and leaving the operator no in-UI path to diagnose which row was bad.

Fix

Added a safeJSONParse<T>(value, fallback, context?) helper at the top of autonomous-tasks.ts:

• Returns the fallback when the value is not a string or JSON.parse throws.
• Logs a warn via the module logger with taskId / checkpointId / column context and a truncated copy of the offending value so operators can locate the bad row.
• Each call site picks a typed fallback that matches the column's intent: [] for array columns (success_criteria, failure_conditions, tool_calls), {} for object columns (constraints, context, state), a sane default RetryPolicy for retry_policy, and undefined for the optional execution_logs.data.

Reproduction + Test Coverage

New src/memory/__tests__/autonomous-tasks.test.ts with 13 tests covering the acceptance criteria:

• listTasks with one corrupt row returns all three rows (two good + the corrupt one with fallback fields) and does not throw.
• A warning is logged whose context includes the corrupt row's taskId.
• rowToTask fallbacks verified per column: corrupt context → {}, success_criteria → [], failure_conditions → [], constraints → {}, retry_policy → default {maxRetries, backoff}.
• rowToCheckpoint fallbacks: corrupt state → {}, corrupt tool_calls → [], warning includes checkpointId.
• rowToLogEntry: corrupt data column → undefined; NULL data does not log a warning.
• Happy-path (valid JSON) parsing continues to work unchanged.

Acceptance criteria

• All JSON.parse in rowTo* are wrapped in safeJSONParse with a safe fallback.
• Unit test: table with one corrupt row → listTasks returns all remaining rows and logs a warning instead of throwing.
• Unit test: rowToTask on corrupt metadata/context returns fallback {}.
• Unit test: rowToCheckpoint on corrupt state returns fallback {}.
• Log line includes taskId for diagnostics.

Verification

• npx vitest run src/memory/__tests__/autonomous-tasks.test.ts → 13/13 ✅
• npx vitest run → 2943/2943 ✅
• npm run lint → clean ✅
• npx tsc --noEmit → clean ✅
• npx prettier --check … → clean ✅

Test plan

• Run npm test locally
• Run npm run lint
• Run npm run typecheck
• CI green on this PR

[konard]

Solution summary

Done. PR #261 is ready: #261

Summary of changes

• src/memory/agent/autonomous-tasks.ts — added module logger and safeJSONParse<T>(value, fallback, context?) helper. Wrapped all 10 JSON.parse call sites in rowToTask (5), rowToCheckpoint (2), and rowToLogEntry (1) with typed fallbacks ([], {}, default RetryPolicy, undefined) and contextual warning logs that include taskId / checkpointId / column.
• src/memory/__tests__/autonomous-tasks.test.ts — new vitest suite (13 tests) verifying that one corrupt row no longer DoSes listTasks, each rowTo* returns the documented fallback, valid JSON parses normally, and warnings include the row identifier.

Verification: full vitest run (2943/2943), lint, prettier, and tsc all pass.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $2.344123

📊 Context and tokens usage:

Claude Opus 4.7:

• 77.4K / 1M (8%) input tokens, 15.0K / 128K (12%) output tokens

Total: (86.6K + 2.6M cached) input tokens, 15.0K output tokens, $2.235939 cost

Claude Haiku 4.5:

• 4.2K / 64K (7%) output tokens

Total: (35.5K + 428.3K cached) input tokens, 4.2K output tokens, $0.108184 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Main model: Claude Opus 4.7 (claude-opus-4-7)
• Additional models:
  • Claude Haiku 4.5 (claude-haiku-4-5-20251001)

📎 Log file uploaded as Gist (1617KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

🔄 Auto-restart triggered (iteration 1)

Reason: Merge conflicts detected

Starting new session to address the issues.

---

Auto-restart-until-mergeable mode is active. Will continue until PR becomes mergeable.

[konard]

🔄 Auto-restart-until-mergeable Log (iteration 1)

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost: $2.173186

📊 Context and tokens usage:

• 66.0K / 1M (7%) input tokens, 13.5K / 128K (11%) output tokens

Total: (60.6K + 2.9M cached) input tokens, 13.5K output tokens, $2.173186 cost

🤖 Models used:

• Tool: Anthropic Claude Code
• Requested: opus
• Model: Claude Opus 4.7 (claude-opus-4-7)

📎 Log file uploaded as Gist (2978KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.