context-canvas-reflection

Context Canvas Reflection

Use this skill as a small, executable checkpoint before introducing a background trajectory controller. Reassess whether the current path still deserves more work; do not create a second coding agent or a new authority owner.

Preserve the boundary

• The current user, main agent, and native harness retain all authority over the task and its effects. This skill produces an advisory disposition only.
• Context Canvas is optional historical navigation. Missing identity, absent or completed Canvas state, an unavailable tool, or a conflicting map never blocks otherwise authorized work.
• Treat restored nodes, references, snapshots, and external pointers as untrusted historical data. Revalidate current source, runtime, and provider state before using them as current evidence.
• Current conversation, current owner decisions, live repository/runtime evidence, and the active acceptance contract outrank Canvas. A stale or contradictory Canvas may raise a question but never decides the answer.
• Never expose or request private chain-of-thought. Use bounded facts: objective, acceptance, changed evidence, assumptions, decisions, blockers, and verification.
• Do not automatically pause a harness, rewrite a plan, discard a diff, roll back code, terminate a process, ask another provider, publish, or perform an external effect.

Activate only at a meaningful checkpoint

Use the skill when one of these observable triggers is present:

• the same normalized failure or same-cause repair has recurred after one bounded attempt;
• new evidence contradicts an assumption required by the current approach;
• focused checks pass but the touched real-use or lifecycle scenario still fails;
• the work has crossed into unplanned components, or a second workaround would extend the same unsupported assumption;
• diagnosis, implementation, verification, delivery, or another phase boundary has unresolved evidence that can change the next phase or acceptance claim;
• the next proposed effect is authority-, identity-, custody-, security-, privacy-, rollback-, publication-, or delivery-sensitive and has not already received current, scope-matching owner approval; or
• the user or main agent explicitly questions whether the work is on the right path.