reflect
by alirezarezvanireflect is a mid-conversation reassessment skill for pausing execution, rereading from the original goal, checking bias and drift, and deciding whether to continue, pivot, or pause. Use it for Decision Support, planning, debugging loops, and strategy work when you need a frank sanity check instead of another summary.
This skill scores 84/100, which makes it a solid listing candidate for directory users who want a reusable mid-conversation reassessment workflow rather than a generic “think harder” prompt. It is highly triggerable, has a clear intended output discipline, and includes useful references and validation scripts, though adoption would be easier with explicit install guidance and more visible quick-start examples.
- Very clear trigger surface: the frontmatter lists explicit phrases such as “reflect,” “take a step back,” “sanity check this,” plus implicit triggers like long detail-mode conversations or signs of being stuck.
- Operational workflow is specific: pause execution, re-read the conversation, assess macro perspective, gaps, reflective inquiry, bias, and alignment, then end with a continue/pivot/pause recommendation.
- Good supporting material: three focused references define the five-bias canon, reflection practice, and honest-output discipline, while scripts help detect bias patterns, depth triggers, and recommendation quality.
- No install command or README is present in the skill path, so directory users may need the broader repository installation pattern to adopt it.
- Support scripts are explicitly heuristic regex/counting tools; they can surface signals but do not replace the skill’s required judgment over the full conversation.
Overview of reflect skill
What reflect does
reflect is a mid-conversation reassessment skill for Claude that pauses execution, rereads the conversation from the original goal, and gives a frank directional judgment. Instead of continuing deeper into details, the reflect skill checks whether the current path still makes sense, what assumptions are driving it, what may be missing, and whether the next move should be to continue, pivot, or pause.
It is especially useful for Decision Support, planning, product work, debugging loops, strategy discussions, and any long exchange where tactical progress may be hiding strategic drift.
Best-fit users and situations
Install reflect if you often ask an AI assistant to help with multi-step work and want a reliable “sanity check” mechanism. It fits users who say things like “take a step back,” “are we missing something?”, “are we overthinking this?”, “zoom out,” or “are we still on track?”
The strongest use case is not summarization. The real job is directional correction: catching confirmation bias, sunk cost thinking, anchoring, complexity creep, or recency bias before the conversation commits to the wrong path.
What makes reflect different
The skill is intentionally low-intake. If the prior conversation contains enough context, it should run immediately rather than asking for a long setup. If the context is too thin, it asks one forcing clarifier instead of turning reflection into another planning exercise.
Its supporting references add useful discipline: reflection should reread from the original goal, avoid inventing problems just to sound rigorous, and end with a concrete recommendation. The included scripts are also helpful for teams that want to test transcripts or validate whether reflection outputs are specific enough.
How to Use reflect skill
reflect install and repository path
Install the skill from the GitHub repository with:
npx skills add alirezarezvani/claude-skills --skill reflect
The source lives at:
productivity/reflect/skills/reflect
Read SKILL.md first, then inspect these files for the operating model:
references/cognitive_bias_canon.mdreferences/conversation_reflection_practice.mdreferences/honest_output_discipline.mdscripts/bias_pattern_detector.pyscripts/conversation_depth_analyzer.pyscripts/directional_recommendation_validator.py
This is a pure-reasoning skill. It does not require external APIs, databases, or vendor-specific tooling beyond a compatible skill environment.
How to invoke reflect in practice
Use direct language inside an active conversation:
- “Reflect on where we are before we continue.”
- “Take a step back. Are we solving the right problem?”
- “Sanity check this plan for bias and drift.”
- “Are we overthinking this, or is the complexity justified?”
- “Use reflect for Decision Support: should we continue, pivot, or pause?”
A strong reflect usage prompt includes the decision pressure, not just the topic. For example, instead of “reflect on this,” write: “Reflect before we implement. Original goal: reduce onboarding friction. Current path: adding a multi-step configuration wizard. Check whether this is still aligned, what assumptions we are making, and whether we should continue, pivot, or pause.”
Inputs that improve output quality
The skill works best when the conversation already contains the original goal, current plan, constraints, and recent decisions. If those are scattered, restate them briefly before invoking reflect.
Useful input pattern:
- Original goal: what success was supposed to mean
- Current direction: what the conversation is now optimizing for
- Decision point: what you are about to commit to
- Constraints: deadlines, risk tolerance, user needs, technical limits
- Concern: drift, complexity, bias, lack of evidence, or stuckness
This helps the skill compare the current path against the starting frame instead of merely reacting to the last few turns.
Optional script-assisted workflow
The scripts are not required, but they are useful for repeatable review. Save a transcript as plain text with User: and Assistant: turns, then run:
python scripts/conversation_depth_analyzer.py --conversation /tmp/transcript.txt
Use bias_pattern_detector.py to flag candidate bias patterns, then let the AI apply judgment rather than treating regex hits as proof. After generating a reflection, directional_recommendation_validator.py can check whether the output ends with a specific continue, pivot, or pause recommendation and avoids vague reassurance.
reflect skill FAQ
Is reflect just a better summary prompt?
No. A summary compresses what happened; reflect judges whether the conversation is still aimed at the right outcome. The reflect skill should surface drift, missing evidence, overbuilt solutions, neglected constraints, or unjustified momentum. If everything is genuinely sound, it should say so with evidence rather than manufacture objections.
When should I not use reflect?
Do not use it for simple one-shot tasks, factual lookups, or moments where there is no meaningful prior context to reassess. It is also a poor fit when you only want encouragement or stylistic feedback. The skill is designed to interrupt momentum, which is valuable during complex work but unnecessary for trivial tasks.
Is reflect beginner-friendly?
Yes, because invocation can be as simple as “reflect” or “take a step back.” Beginners benefit from the skill’s low-intake design, while advanced users can get better results by giving the original goal, the current decision, and the suspected failure mode.
How does reflect for Decision Support help teams?
For Decision Support, reflect gives a structured pause before commitment. It is useful before choosing a product direction, finalizing a technical design, continuing a debugging path, or accepting a recommendation. Its value is strongest when the cost of continuing in the wrong direction is higher than the cost of pausing for reassessment.
How to Improve reflect skill
Give reflect a clear decision edge
The most common weak output comes from vague invocation. “Reflect on our discussion” can work, but “Reflect before we choose option B; check for anchoring, missing counterevidence, and whether we should continue, pivot, or pause” gives the skill a sharper target.
If you care about a specific risk, name it. For example: “I’m worried we are adding features because each one sounds reasonable individually. Check for complexity bias.”
Force evidence-based reassessment
A good reflect output should cite concrete conversation evidence: the original goal, the point where the framing shifted, the assumption that became dominant, or the decision that needs validation. If the response feels generic, ask: “Redo the reflection using specific evidence from earlier turns and end with a concrete recommendation.”
This aligns with the repository’s honest-output discipline: no vague pessimism, no vague reassurance, and no invented problems.
Watch for common failure modes
Common reflect failures include focusing only on the latest turns, summarizing instead of judging, finding problems without evidence, or ending without a clear next move. Another failure is over-questioning: the skill should ask a clarifier only when the conversation is too thin to reassess.
You can correct these by saying: “Reread from the original goal, not just the recent implementation details. Identify the strongest reason to continue and the strongest reason to pivot.”
Iterate after the first recommendation
Treat the first reflection as a decision checkpoint, not the end of the work. If the recommendation is “continue,” ask what evidence would change that view. If it is “pivot,” ask for the smallest viable pivot. If it is “pause,” ask what exact missing information must be resolved before continuing.
This keeps reflect practical: the goal is not more analysis, but a better next action.