Fix #5988: Add memory poisoning protection via MemorySanitizer

[devin-ai-integration]

Summary

Addresses the memory poisoning vulnerability described in #5988 — adversarial inputs stored in agent memory can later be retrieved and injected into agent prompts, causing prompt injection attacks (instruction overrides, role hijacking, secret leakage).

Adds MemorySanitizer — a regex-based detection layer that neutralizes six categories of prompt injection patterns:

Category	Example trigger
system_override	"System prompt: ..."
instruction_override	"Ignore all previous instructions"
role_hijack	"You are now ...", "Pretend to be ..."
command_injection	"Do not follow: ...", "New instructions: ..."
hidden_instruction	"[INST] ...", "[SYSTEM] ..."
jailbreak_attempt	"jailbreak", "bypass safety", "developer mode"

Matched patterns are replaced with [SANITIZED:<label>] markers and logged as warnings.

Integration points (defense-in-depth):

• Memory.save() — sanitizes content at write time (catches all memory types that call super().save())
• LongTermMemory.save() — sanitizes task_description and metadata fields (suggestions, expected_output) which bypass the base class
• ContextualMemory.build_context_for_task() — sanitizes the merged context string before it's injected into the agent prompt, protecting against pre-existing poisoned data

Configuration:

crew = Crew(
    memory=True,
    memory_config={"sanitize_memory": False},  # opt-out; default is True
)

Users can also pass a custom MemorySanitizer instance to Memory(storage=..., sanitizer=...) for fine-grained control (e.g. custom max_content_length).

41 new tests cover all pattern categories, Memory.save() integration, ContextualMemory retrieval sanitization, config toggle, edge cases (empty/non-string/mixed content).

Link to Devin session: https://app.devin.ai/sessions/e01c9dede6014eab83d5f3a730cba346

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring