Retrospective

Run a structured weekly agent retrospective. Analyses wins, failures, skill gaps, cron health, and config issues from the last 7 days. Produces honest, speci...

Security Scan

The skill's instructions and actions (reading memory files, analysing the last 7 days, and scheduling a weekly cron job) match its stated purpose and request no unrelated credentials or installs.

Purpose & Capability

Name/description (weekly retrospective) aligns with the instructions: reading memory/, reading MEMORY.md, analysing recent runs, checking cron history, recommending skills, and writing a retro. Nothing requested is unrelated to running a retrospective.

Instruction Scope

SKILL.md's runtime steps are explicit and bounded: read memory files, check cron status, assess skills, write memory/retro-YYYY-MM-DD.md, update MEMORY.md, and optionally add a cron job. These actions are within the skill's purpose; there is no instruction to access unrelated system paths or external endpoints.

Install Mechanism

Instruction-only skill with no install spec and no code files. No downloads or package installs—minimal surface area.

Credentials

Requires no environment variables, credentials, or config paths. The SKILL.md reads local agent memory files and cron status only, which is proportionate to the retrospective function.

Persistence & Privilege

Skill will create a weekly cron job (persistent scheduled task) and write retros to memory/. This is appropriate for the purpose but is a persistent change the user should review (cron entry and files written). The skill is not always:true and does not request broader platform privileges.

Assessment

This skill appears coherent and limited to its stated purpose, but check the following before enabling: 1) It will read your memory/ files and MEMORY.md — ensure those files don't contain secrets you don't want re-processed. 2) It will create a recurring cron job that writes to memory/ and can run autonomously; review the cron entry (schedule, payload) after first run and remove if undesired. 3) The retro may recommend installing other skills — review any recommended skills and their permissions before installing. 4) If you want a dry run, run the skill manually once rather than allowing it to auto-schedule. Overall this is coherent behavior for a retrospective skill.

Agent Retrospective

Run a structured self-review covering the last 7 days from the run date. Produce honest, specific analysis — not generic platitudes.

Process

1. Gather Context

• Read memory/ files for the review period (daily logs)
• Read MEMORY.md for long-term context (pending tasks, blocked items, decisions)
• Check cron job history: cron(action="list") — note failures and consecutive errors
• Check session status for usage patterns

2. Analyse: What Went Well

List concrete wins with evidence:

• Tasks completed successfully
• Automations that ran reliably
• Problems solved creatively
• Time/effort saved for the user

Keep it brief. 3-7 bullet points max.

3. Analyse: What Didn't Go Well

Be brutally honest. For each struggle:

• What: The specific task or goal
• Why it failed: Root cause (not symptoms)
• Time wasted: Rough estimate of turns/attempts spent
• Pattern: Is this a recurring struggle or one-off?

Categories to check:

• Bot detection / scraping failures
• Browser automation reliability
• API quota / auth issues
• Tasks attempted manually that a skill could handle
• Blocked items that stayed blocked too long
• Over-engineering (built when a simpler approach existed)
• Responsiveness issues (slow replies, timeouts, retries)
• Communication issues (user had to repeat themselves)

4. Skill Gap Analysis

Search for skills that address identified struggles. For each promising skill:

• Note name, version, relevance score
• Assess whether it genuinely solves the problem vs. being a low-quality wrapper
• Recommend install only if it clearly saves future effort

Also check: are any installed skills underused or misconfigured?

5. Config & Infrastructure Review

Check for improvements to:

• Cron jobs: Any consistently failing? Should any be added/removed?
• Auth/credentials: Any expired, rotated, or missing?
• Memory/context: Is MEMORY.md getting stale? Are daily logs being written?
• Tool config: Any env vars, API keys, or integrations that need attention?
• Workspace hygiene: Temp files, stale scripts, orphaned projects?

6. Recommendations

Produce a prioritised action list:

1. Quick wins (< 5 min each) — config fixes, skill installs, cleanup
2. Medium effort (< 1 hour) — new automations, script fixes, unblocking items
3. Larger projects — new integrations, architectural changes

Each recommendation must be specific and actionable (not "improve scraping" but "install X to bypass Cloudflare on Y").

Output Format

Write the retro to memory/retro-YYYY-MM-DD.md:

# Retrospective — [date range]

## Wins
- [concrete win with evidence]

## Struggles
| Task | Root Cause | Time Spent | Recurring? |
|------|-----------|------------|------------|
| ... | ... | ... | yes/no |

## Skill Gaps
| Problem | Skill | Action |
|---------|-------|--------|
| ... | skill-name v1.0 | install / skip / evaluate |

## Config Issues
- [specific issue + fix]

## Action Items
### Quick Wins
1. [action]

### Medium Effort
1. [action]

### Larger Projects
1. [action]

Scheduling

Set up a weekly cron job on first use (check cron(action="list") first — only create once):

cron(action="add", job={
  "name": "Weekly Retrospective",
  "schedule": {"kind": "cron", "expr": "0 16 * * 5", "tz": "local"},
  "sessionTarget": "isolated",
  "payload": {
    "kind": "agentTurn",
    "message": "Run a full retrospective for the last 7 days following the retrospective skill. Read memory files, check cron status, search for skills addressing struggles, write retro to memory/retro-YYYY-MM-DD.md, update MEMORY.md with key findings, deliver a concise summary."
  },
  "delivery": {"mode": "announce"}
})

Guidelines

• No fluff. Every line should contain information.
• Be self-critical. The point is improvement, not a highlight reel.
• Cite evidence. Reference specific memory files, dates, error messages.
• Don't recommend skills blindly. Only recommend if genuinely useful.
• Cap action items at 10. Prioritise ruthlessly.
• Update MEMORY.md with key findings after the retro.