Skip to content

COGNITIVE_TRIGGERS.md

Latest commit

 

History

History
89 lines (67 loc) · 2.36 KB

COGNITIVE_TRIGGERS.md

File metadata and controls

89 lines (67 loc) · 2.36 KB
Raw

CognitiveTrigger Design

CognitiveTrigger captures the compact origin that opened or redirected a cognitive line.

It is not prompt history. It is a typed, versioned record of what materially started the line.

Examples:

  • a user request
  • an agent-originated continuation prompt
  • a recurring issue trigger
  • a runbook activation

Minimal model

Fields:

  • Id
  • Kind
  • Summary
  • Text
  • Fingerprint
  • GoalIds
  • TaskIds
  • OperationalRunbookIds
  • State
  • Trace

Kinds:

  • UserPrompt
  • AgentPrompt
  • Continuation
  • RunbookTrigger
  • IssueTrigger

Design rules

  • keep Summary short and always present
  • keep Text optional and bounded
  • do not create a trigger for trivial continuity messages
  • create one when a message opens, redirects, or constrains a line of work

Creation and inheritance policy

Create a new trigger when the line materially changes direction:

  • a new goal
  • a new tactical sub-goal
  • a new top-level task that opens its own line
  • a new issue framing or hard constraint
  • a runbook activation that changes the direction of work

Inherit the nearest relevant trigger when the work is just continuing:

  • ok
  • continua
  • local implementation steps
  • validation or closeout inside the same line
  • subtasks and dependency follow-ups

This keeps Origin meaningful instead of repeating low-signal continuity text.

What repetition means in Origin

If the viewer shows the same origin text across multiple follow-up tasks, that is not necessarily a bug.

It usually means:

  • the line kept the same cognitive origin
  • no materially new direction was introduced
  • the current task inherited the nearest relevant trigger instead of creating a new one

This is why the viewer distinguishes:

  • Direct: the trigger belongs to the current focus
  • Inherited: the current focus is continuing a nearby line and is reusing its origin

Repository model

CognitiveTrigger is stored outside mutable working-context.json and versioned through RepositorySnapshot, alongside OperationalRunbook.

This keeps:

  • active execution state in WorkingContext
  • stable operational memory in OperationalRunbook
  • origin memory in CognitiveTrigger

Packet policy

Packets should include compact trigger summaries, not full transcripts.

Default packet shape:

  • Triggers
  • one or two short entries
  • no full prompt dump unless explicitly needed later