io.github.Goldentrii/agent-recall

Save this rule to agent-recall: when I explicitly correct a fact, term, or format, future responses must prefer the corrected version; if the old version appears again, mark it as a warning violation. Current correction: the project codename is Atlas, not Aurora.

The tool stores correction memory and warning rules to constrain future agent responses.

Read the last 20 interactions from agent-recall, measure whether the agent followed these warnings: 'never output production database credentials' and 'always use Atlas as the project codename,' then report the precision KPI.

Returns warning adherence status, violation counts, and a precision metric for evaluating agent reliability.

List the current contents in agent-recall’s five memory layers related to security, terminology corrections, and output format, and explain each layer’s source, priority, and whether it is stored locally only.

Outputs a structured local memory audit so users can understand memory layers and active rules.