Skip to content
Back to skills

Generate Reflective Note

ASecurity

Generate content with awareness of the agent's current reasoning state — goal analysis, situation context, character orientation. Like generate-note but informed by internal planner state.

  • 10 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added June 6, 2026
ai-agentspythongo

Security analysis

A100/100

Pro scans all 2 files and shows the line behind each finding

Scanned June 6, 2026

npx -y skills add bdambrosio/Cognitive_workbench --skill generate-reflective-note --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Generate Reflective Note?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Generate Reflective Note
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/bdambrosio-generate-reflective-note/badge)](https://www.skillsdirectory.com/skills/bdambrosio-generate-reflective-note)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: generate-reflective-note
type: python
description: "Generate content with awareness of the agent's current reasoning state — goal analysis, situation context, character orientation. Like generate-note but informed by internal planner state."
---

# generate-reflective-note

LLM-based content generation with access to the agent's internal reasoning state.
Like `generate-note`, but the LLM prompt includes the planner's Stage 1 context
(character orientation, goal analysis, tool selection, situation awareness) as
background, enabling state-aware or reflective generation.

## Input

- `prompt`: Generation instruction (required)
- `style`: `"code"` or `"text"` (optional, default: `"text"`)
- `target_tokens`: Integer (optional). Target output length in tokens. Overrides default max_tokens when provided via OUTPUT GUIDANCE.

## Output

Success (`status: "success"`):
- `value`: Generated content informed by agent state

Failure (`status: "failed"`):
- `reason`: Error description

## Behavior

- Automatically retrieves the planner's reasoning snapshot (if available)
- Includes it as `## Internal Reflective State` context in the LLM prompt
- The prompt is the actual query; the reflective state is background
- Falls back to regular generation if no planner state is available

## Planning Notes

- Use when the agent needs to generate content that reflects its current understanding,
  orientation, or reasoning about the goal and situation
- Use `generate-note` for plain generation without agent state awareness
- Typical uses: self-assessments, progress reflections, situation reports,
  agent-perspective summaries, epistemic status notes

## Examples

```json
{"type":"generate-reflective-note","prompt":"Assess your current understanding of the user's goals and any gaps","out":"$assessment"}
{"type":"generate-reflective-note","prompt":"What aspects of this task are you most uncertain about?","out":"$uncertainty"}
{"type":"generate-reflective-note","prompt":"Summarize what you know about the current situation from the agent's perspective","out":"$sitrep"}
```

Files in this skill

  • Skill.md2 KB
  • tool.py5.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.