self-reflect

name	self_reflect
version	1.0.0
description	Extract grounded lessons learned from a completed task and optionally write reusable experience entries into memory. Primarily used by the runtime or agent as a post-task reflection step.
invocation	blue self_reflect goal="Fix parser regression" final_status=completed result_summary="Patched parser and passed focused verification" --json
examples	["blue self_reflect goal=\"Fix parser regression\" final_status=completed result_summary=\"Patched parser and passed focused verification\" --json","blue self_reflect goal=\"Investigate flaky test\" final_status=failed failure_reason=\"Timeout in CI\" --json"]
capability_tags	["reflection","lessons","memory"]
interaction_mode	stateless
card_support	none
category	internal
tags	["reflection","memory","internal"]

Self Reflect

Summarize a finished task, extract grounded reusable lessons, and persist them into memory when that backend is available.

This skill is mostly for runtime-driven post-task reflection. It is not usually the first choice for ordinary end-user requests.

Setup

No external dependencies required. This skill needs the runtime self-reflection backend to be wired; otherwise execution returns a backend-unavailable error.

---

Task Routing

User Intent	Action
Agent/runtime wants an automatic post-task reflection pass	blue self_reflect ...
User explicitly asks to review lessons learned from a completed task	blue self_reflect ...
Need reusable memory-worthy lessons rather than a normal prose recap	blue self_reflect ...
User only wants a plain summary or retrospective in chat	Prefer a normal answer unless structured reflection output is needed

---

Command Usage

Basic successful reflection:

blue self_reflect goal="Fix parser regression" final_status=completed result_summary="Patched parser and passed focused verification" --json

Failure reflection:

blue self_reflect goal="Investigate flaky test" final_status=failed failure_reason="Timeout in CI" --json

With plan and verification context:

blue self_reflect goal="Refactor scheduler docs" final_status=completed result_summary="Updated docs and tests" verification_output="go test ./internal/skillmanifest passed" --json

Advanced proposal/review mode is also supported through additional fields such as source_kind, source_id, evaluation_summary, proposal_candidates, and proposal_mode.

---

Parameters

• goal: Required for normal reflection flows.
• task_id: Optional task identifier for memory tagging.
• plan: Optional step list with description, status, and output.
• step_outputs: Optional step outputs merged into plan by index.
• verification_output: Optional verification stage output.
• final_status: Required for normal reflection flows; must be completed, failed, or partial.
• result_summary: Optional short outcome summary.
• failure_reason: Optional failure reason.
• source_kind, source_id, evaluation_summary, proposal_candidates, proposal_mode: Optional advanced proposal inputs used by internal review workflows.

---

Output

Returns structured JSON including:

• summary
• lessons
• memory_written
• skipped_reason
• proposal_count
• proposal_ids
• proposal_skipped_reason

---

Error Handling

Error	Resolution
Missing goal for a normal reflection flow	Provide goal, or use proposal inputs if running proposal-only mode
Missing or invalid final_status	Use one of completed, failed, or partial
Self-reflection backend not available	Run in a runtime where the reflection backend is wired

---

Notes

• "When available" means the reflection backend can write into the configured memory layer only when that memory service is present in the current runtime.
• In normal agent flows this skill is often invoked automatically after task completion rather than explicitly requested by the user.
• Use this skill for structured lessons extraction, not as the default way to write a normal summary for the user.

Example Triggers

• "帮我复盘一下这次任务"
• "Extract lessons learned from this task"
• "总结一下这次失败里能复用的经验"