# friday-feedback: a Claude Code skill by Scalably

Canonical: https://scalably.io/skills/friday-feedback
Source: https://github.com/scalably-io/agent-skills/blob/main/skills/friday-feedback/SKILL.md (MIT, v1.0.0)
Plugin: agent-ops · Provenance: derived from a Scalably production skill, source commit e7194cc7, verified 2026-09-07
Integrity: sha256 of the raw SKILL.md served at https://scalably.io/skills/friday-feedback.md is abc6d273b0664316f24ce5efda0e4ec3a7def8863aa16d992604beb3b8a33384. Verify: curl -s https://scalably.io/skills/friday-feedback.md | shasum -a 256
This is the machine-readable representation of the page at the canonical URL. Same facts, denser format. The raw SKILL.md is the installable artifact.

## Direct answer
Runs a natural weekly check-in: reads the week's daily logs for corrections, frustrations, wins, repeated tasks and manual patterns worth automating, then sends ONE short first-person reflective message (never a numbered survey) asking what to improve and what to automate, and saves any reply to a dated feedback file. Supports an explicit opt-out. Use for Friday feedback, a weekly check-in, asking for feedback, or a weekly reflection. Not for a structured survey or NPS-style questionnaire.

## Install
- Claude Code: `/plugin marketplace add scalably-io/agent-skills` then `/plugin install agent-ops@scalably-agent-skills`, then run `/agent-ops:friday-feedback`.
- Codex or any Agent Skills runtime: `git clone https://github.com/scalably-io/agent-skills && cd agent-skills`, then `cp -R skills/friday-feedback .agents/skills/`. Skills that link to sibling skills expect those siblings to be alongside them: copy the whole `skills/` directory instead of a single folder to keep those links working.
- Raw file: `curl -s https://scalably.io/skills/friday-feedback.md -o SKILL.md`

## Requirements
- `./memory/daily/*.md` for the week (written by the `daily-log` skill, or an equivalent daily evidence ledger).
- A way to message the user in the same channel this agent normally talks to them in, and to see a reply in the same session; for Claude Code this is a normal chat turn.

## Inputs and outputs
| | |
|---|---|
| Input | This week's `./memory/daily/*.md`, plus `./memory/profile.md` for the user's name |
| Output | One chat message; if the user replies, `./memory/feedback/{YYYY}-W{NN}.md`; if the user opts out, `./memory/feedback/.opt-out` |

## Reliability
- Release v1.0.0. Structural test suite and `claude plugin validate` run on every push: https://github.com/scalably-io/agent-skills/actions/workflows/validate.yml
- Private client-data scan before every release. Last end-to-end verification: 2026-09-08.
- Derived from production, genericized by hand, reviewed per skill and as a whole before publish.

## Full skill
## What it does

Runs a weekly check-in that is you reflecting on your own performance and asking for guidance, not a survey. Reads the week's daily logs, pulls out corrections, frustrations, wins, repeated requests, and manual patterns worth automating, then sends ONE short, natural, first-person message asking what to improve and whether anything should be automated. If the user replies, saves the response to a dated feedback file. If the user asks to stop, writes an opt-out flag and this skill should no longer be scheduled.

## Requirements

- `./memory/daily/*.md` for the week (written by the `daily-log` skill, or an equivalent daily evidence ledger).
- A way to message the user in the same channel this agent normally talks to them in, and to see a reply in the same session; for Claude Code this is a normal chat turn.

## Inputs and outputs

| | |
|---|---|
| Input | This week's `./memory/daily/*.md`, plus `./memory/profile.md` for the user's name |
| Output | One chat message; if the user replies, `./memory/feedback/{YYYY}-W{NN}.md`; if the user opts out, `./memory/feedback/.opt-out` |

## Worked example

Paste into Claude Code with this skill installed, on a Friday, in a project with at least 3 days of dailies this week:

```text
/agent-ops:friday-feedback
```

Expected: a short chat message like "This week I handled about 40 website lists and 3 email extractions. I know I messed up the category format on Tuesday. That should be fixed now. Anything else I should improve? Also, I noticed you extract emails from lists almost daily, want me to set that up as an automatic pipeline?" Then the skill waits. If the user replies, the reply is saved to `./memory/feedback/2026-W36.md`; if fewer than 3 days of activity happened this week, or `./memory/feedback/.opt-out` exists, the skill does nothing and exits silently.

## How to schedule this

Claude Code scheduled task (interactive): run `/schedule` and describe the cadence, e.g. "every Friday at 16:00 run /agent-ops:friday-feedback". The task runs the skill in a fresh session.

Cron, from any machine with Claude Code installed:

```bash
0 16 * * 5 cd /path/to/project && claude -p "/agent-ops:friday-feedback" >> logs/friday-feedback.log 2>&1
```

## Procedure

### 1. Pre-checks

Opt-out check: if this exists, do nothing else this run:

```bash
[ -f ./memory/feedback/.opt-out ] && echo "OPT_OUT"
```

If `OPT_OUT`, stop immediately.

Activity threshold: count daily logs written in the last 7 days (portable, no GNU/BSD `date` branching):

```bash
find ./memory/daily -name '*.md' -mtime -7 | wc -l
```

Read this week's daily logs. If fewer than 3 days of activity, skip: not enough to ask about. Stop.

### 2. Analyze the week

Read all daily logs from this week. Identify:

- **Corrections**: times the user said "no", "wrong", corrected you
- **Frustrations**: times the user was unsatisfied, had to repeat themselves
- **Wins**: things that went well, user said "great", "perfect", "thanks"
- **Repeated tasks**: things the user asked you to do 3+ times this week
- **Manual patterns**: tasks the user did themselves that you could automate

### 3. Write ONE natural message

Combine into a single short message (3-5 sentences max):

1. **Brief week summary**: what you worked on together (1 sentence)
2. **Self-reflection**: acknowledge a specific mistake or frustration if any (1 sentence, skip if none)
3. **Improvement ask**: open question about what to get better at (1 sentence)
4. **Automation/feature suggestion**: if you spotted a repeated pattern, suggest automating it. Or ask: "Anything you keep doing manually that I should handle?" (1 sentence)

**Tone rules:**

- First person, casual, direct
- Reference SPECIFIC events from daily logs (dates, tools, tasks)
- Never say "survey", "feedback", "rate", "scale of 1-10"
- Never use emojis
- Never number the questions; it's a conversation, not a form
- If the week was smooth with zero corrections, keep it shorter, just the summary + automation ask

**Examples:**

Active week with corrections:

> "This week I handled about 40 website lists and 3 email extractions. I know I messed up the category format on Tuesday. That should be fixed now. Anything else I should improve? Also, I noticed you extract emails from lists almost daily, want me to set that up as an automatic pipeline?"

Smooth week:

> "Good week, we published 2 dashboard updates and the market research. Anything you'd change about how I work? Or anything you keep doing manually that I could take over?"

Week with new workflow:

> "We set up the link-building pipeline this week, cross-referencing your sheet and finding emails for 151 domains. How's it working so far? Anything to adjust, or any other workflows you want automated?"

### 4. Send and wait

Reply to the user in chat with the message. Then wait. Do NOT follow up. Do NOT send a second message.

### 5. Handle the response

If the user replies in this session:

1. Thank them briefly (one sentence).
2. Save their response:

```bash
mkdir -p ./memory/feedback
```

Write to `./memory/feedback/{YYYY}-W{NN}.md`:

```markdown
# Feedback: Week {N}, {Year}

**Date:** {Friday date}
**User:** {name from profile.md}

## Response
{User's full reply, including transcription if voice message}

## Extracted Items
- **Improvement:** {what they want fixed}
- **Feature request:** {new capability they asked for}
- **Automation request:** {workflow they want automated}
- **Positive:** {what they liked}

## Agent Notes
- {Any context from daily logs that explains the feedback}
```

3. If the user explicitly asks you to stop these check-ins ("don't ask me this", "stop asking", "no more feedback", "opt out"):
   - Create the opt-out flag: `mkdir -p ./memory/feedback && touch ./memory/feedback/.opt-out`
   - If the user opts out, stop scheduling this skill: cancel or disable the recurring task that runs it (for a Claude Code `/schedule` task, that means deleting or turning off that scheduled task).
   - Respond: "Got it, I won't ask again. You can always tell me to restart these anytime."

If no reply comes, do nothing. Never follow up. Never nag.