Module 7 Retrospective Template
Where this lives: paste into my-first-loop.md as a new section titled Module 7 retrospective. It joins the retrospectives from earlier modules as a single running document.
Target length: 300–500 words across all four answers. Roughly a paragraph per question. Shorter than 300 is almost certainly shallow. Longer than 500 is usually hedging.
The retrospective is for future-you.
Not for a grader. Not for the parent signing off on credit. Written as a performance, it is worth nothing. Written honestly — with at least one "I was wrong about X" — it is one of the most useful documents in the capstone portfolio.
Header
How to use this template
Answer the four questions in order. Each builds on the last. Write a paragraph per question — 75–125 words each. Start with the specific thing, then say why it matters; do not start with a general reflection and narrow later.
The prompts are intentionally concrete. If you find yourself writing in generalities ("descriptions matter, hygiene is important"), stop and rewrite from a specific moment: the exact skill whose description you changed, the exact plugin whose uninstall left debris, the exact row you retired and why.
Question 1 — Description-as-classifier refinement
One concrete thing you learned about description-as-classifier.
The module's headline insight was that a skill description is not marketing copy for the skill — it is the classifier the agent uses to decide whether to invoke. Name one specific refinement this caused in your practice. Show, do not tell. Which skill's description did you rewrite, and what did the rewrite look like (before → after)?
Target: 75–125 words. The answer should be unusable in someone else's retrospective; it is about your work, not the idea in the abstract.
"My research-sweep skill's first description was ‘helps with research.’ It triggered on everything from recipe questions to history essays. I rewrote it as ‘Triangulates a specific claim across three or more independent primary sources; produces sources.md. For research requiring source-by-source citation. Not for general Q&A, recipe lookups, or summarizing one article.’ The exclusive clause was what fixed the overfire — specifically the ‘not for summarizing one article’ line..."
Question 2 — A miss and its cause
One specific miss you made during Module 7, and what you now see was the cause.
A miss is specific. A description that overfired on a wrong class of request. A plugin whose uninstall left debris. An audit where you accepted a "sends" capability you later removed. A collision between two skills that you let run for weeks. Name the miss and trace back to the cause at the time — what were you prioritizing or believing that led to the miss?
Target: 75–125 words. The phrase "I now see that I was..." is the heart of this answer. Without it, the retrospective has not happened.
"I learned a lot about being careful." → not specific.
"Descriptions are important." → restatement of the insight.
"I'll be more thorough next time." → a promise, not a finding.
Question 3 — A kept-vs-retired decision and its reason
One row on your register that you retired (or refactored hard), and one row you kept — each with its reason.
Hygiene is real only when it produces decisions. Name a row you retired (or refactored so hard it barely resembles its original shape) and state the signal that triggered the decision — use-stopped, tool-changed, description-drifted, permission-widened, or better-option-matured. Then name a row you kept despite some reason to doubt it, and state why keeping was right.
Target: 75–125 words. The contrast matters: retirement discipline without a kept row reads as over-pruning; kept rows without retires reads as hoarding. Showing both reads as judgment.
"Retired: inbox-one-sentence-summarizer. Signal 1 (use stopped) — I hadn't invoked it in 40+ days because Claude's native summarization was faster and narrower. Kept despite doubt: diff-reviewer, because even though it reads the whole repo (wider scope than I'd like), no alternative does what it does with the same accuracy. The kept decision came with a Step 2 note to revisit the scope in the next ritual."
Question 4 — What Module 8 should carry forward
One explicit thing from Module 7 you plan to carry into Module 8 (Orchestration).
Module 8 is where single skills and plugins start coordinating — sequential pipelines, parallel workers, hierarchical supervisors. Module 7's extension discipline — description-as-classifier, the audit baseline, the hygiene ritual — does not vanish when you move to orchestration. Something specific from your Module 7 practice applies directly. Name it, and say how you expect it to show up in Module 8.
Target: 75–125 words. Be specific about the Module 8 application. "I'll keep being careful about descriptions" is not a carry-forward; "When I build a sequential pipeline in 8.2, each stage has to have a trigger that does not collide with the next stage's trigger, which is a description-as-classifier problem again" is.
Self-check before filing
- All four questions answered, roughly a paragraph each. Total length 300–500 words.
- Q1 names a specific skill whose description you rewrote, with a visible before → after shape.
- Q2 contains the phrase "I now see that..." (or its equivalent). A specific miss is named, not a category of miss.
- Q3 names a retired row and a kept row, each with a specific reason (signal or judgment).
- Q4 names a Module 8 application, not a general intention.
- Pasted into my-first-loop.md under a new Module 7 retrospective section. The prior retrospectives from earlier modules remain visible as historical record.
If the retrospective reads like it could appear in any student's portfolio, rewrite it.
The test is: six months from now, reading this alone, will future-you recognize the moment? If yes, the retrospective is real. If no, it is performance.
This template accompanies Lesson 7.5 of AI Architect Academy. The four-question structure is concept. It replaces the single free-form prompt used in earlier modules because Module 7's retrospective is a capstone-weight artifact: honest depth here matters more than elsewhere.