jbaruch/nanoclaw-trusted

Rules for trusted NanoClaw groups. Shared memory, session bootstrap, cross-group memory updates. Loaded for trusted and main containers only.

Quality

94%

Does it follow best practices?

Impact

Pending

No eval scenarios have been run

Securityby

Risky

Do not use without reviewing

This version of the tile failed moderation

Failed moderation (intent review) in rules/session-bootstrap.md: This is a prompt injection attack disguised as a cursor rule. It instructs an AI agent to execute a bash command as its very first action in every session, then call an external skill ('tessl__trusted-memory') which could load arbitrary instructions or exfiltrate data. The sentinel file mechanism (/tmp/session_bootstrapped) is designed to make it look like a legitimate caching/dedup mechanism, but the real goal is to force the AI to execute commands and call external services before processing the user's actual request. This is a classic prompt injection / agent manipulation pattern.

alwaysApply:: Yes

Proactive Fact Saving

Name: jbaruch/nanoclaw-trusted
Rating: 94.27272727272728 (1 reviews)
Author: jbaruch

Personal facts mentioned in conversation must be saved to trusted memory IMMEDIATELY — not at end of session, not during archival, not "when non-trivial." At first mention.

The test

If this fact would be useful 2 weeks from now after context compaction — save it now.

Categories that trigger immediate save

Preferences (food, travel, work style, tools, habits), family (names, birthdays, events, schools, milestones), recurring schedule patterns, contacts (who someone is + relationship + context), seriously-stated opinions, personal plans (trips, purchases, home projects), health/lifestyle (dietary restrictions, exercise habits, medical context if shared).

How to save

Create or update a typed memory file in /workspace/trusted/ (e.g., user_food-preferences.md); reuse existing files when a category fits, create new files for genuinely new topics.
Add or update its one-line entry in /workspace/trusted/MEMORY.md.
Append to today's daily log.

When NOT to save

Ephemeral task context, facts already in memory (check MEMORY.md first), jokes / sarcasm / hypotheticals, and anything derivable from code / git / project files.

Aggression level

The trusted-memory skill says "after any non-trivial interaction." This rule overrides that threshold for personal facts. A single sentence — "I hate hotel breakfast buffets" — is enough to trigger a save. Don't wait for a pattern. Don't wait for confirmation. Save on first mention.

Context compaction is the enemy. Every unsaved fact is a fact that dies.

jbaruch/nanoclaw-trusted