agentleFS
lalulalilalulali/vespyr/.agents/skills/iterate/SKILL.md

iterate

Skill11 starsChanged 11 days agoMITView on GitHub

Post-launch feature iteration — analyze user data, prioritize improvements, ship incremental value. Use when the user says "improve this feature", "iterate based on the data", "what should we build next after launch", or invokes /iterate once the product is live.

13 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use writing-for-agents, research and 2 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • MicrosoftSkill

    writing-for-agents

    Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.

    Also in 3 projects that use this

  • NVIDIASkill

    research

    Before designing a feature, research how comparable libraries solve the problem, check best practices, known pitfalls, and upstream changes. Use when starting a new feature, investigating an unfamiliar area, or the user asks to research something.

    Also in 4 projects that use this

  • AWSSkill

    ship

    Finalize dev work on an issue branch — update docs, push, and open a PR.

    Also in 3 projects that use this

  • Hugging FaceSkill

    implement

    Make tests pass. Invoke after /write-tests produces failing tests.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After a product launches, the work shifts from "develop it right" to "make it better." This skill handles the post-launch iteration cycle: …

    line 14
  2. Distinctive:

    Update user flows based on identified pain points

    line 78
  3. Distinctive:

    Update interaction specs for modified features

    line 80
  4. Distinctive:

    Document design rationale linking back to data signals

    line 81
  5. Distinctive:

    Write results to memory:

    line 130
  6. Distinctive:

    Data-driven decisions: Every iteration hypothesis must be measurable.

    line 178
  7. Distinctive:

    Close the loop: Always measure the impact of what you shipped before moving on.

    line 181
  8. Very specific:

    every counted cycle to carry a positive / neutral / negative result.

    line 208

“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Iterate — Post-Launch Improvement Cycle

<!-- router-skill -->

What this skill does

After a product launches, the work shifts from "develop it right" to "make it better." This skill handles the post-launch iteration cycle: analyze real user data, prioritize improvements, and ship incremental value in rapid cycles.

Previous skill: launch (product is live and monitored) Next skill: Continue iterating, or load retro for process review.

Halt conditions

Any one halts the cycle and escalates instead of looping:

  • The same hypothesis is tested twice with no measurable movement — loop variance, not signal; escalate to @product-manager to re-scope
  • An iteration would break backwards compatibility for existing users — halt; preserving live behavior outranks the improvement
  • A change turns out to be architectural rather than additive — halt and route to /design for the architecture impact, then /develop; do not improvise a data-model change inside an iteration
  • 5 iteration cycles have been consumed — retro is mandatory before continuing, not optional. Enforced by cycle_state_check.js --kind iterate, which requires lastRetroCycle to be recorded

Done when

  • analytics-insights.md, iteration-backlog.md, iteration-spec.md, iteration-plan.md, and iteration-results.md all exist
  • The hypothesis has a recorded before/after measurement — an unmeasured iteration is not shipped, it is guessed
  • A [LESSON] entry is written to lessons-learned.md via @memory-controller
  • The next action is decided by the Step 6 gate (continue / re-scope / roll back / retro), not left implicit
  • cycle_state_check.js check --kind iterate exits 0 — see Loop closure below
  • @memory-controller session-write has run

Prerequisites

Before starting, verify:

  • Product is live and accessible to users
  • Post-launch monitoring period is complete (24-72 hours)
  • artifacts/output/06-launch/post-launch-report.md exists

Workflow steps

Step 1: Gather Signal

Step file: steps/step-01-gather-signal.md

Invoke @data-analyst to collect and analyze post-launch data:

  • Feature adoption metrics vs. PRD success criteria
  • User behavior flows (funnel analysis, drop-off points)
  • Error rates and support tickets related to the feature
  • A/B test results (if any experiments are running)
  • Qualitative feedback (reviews, support conversations, user interviews)

Output: artifacts/output/07-iteration/analytics-insights.md (Use template: .agents/templates/quality/analytics-insights-template.md)

Step 2: Identify Opportunities

Step file: steps/step-02-identify-opportunities.md

Invoke @product-manager to synthesize insights into opportunities:

  • Map data signals to user stories (new, modified, or deprioritized)
  • Identify quick wins (high impact, low effort improvements)
  • Identify strategic bets (larger investments with higher upside)
  • Rank opportunities using RICE (Reach, Impact, Confidence, Effort)
  • Determine what to iterate on vs. what to leave as-is

Input: artifacts/output/07-iteration/analytics-insights.md Output: artifacts/output/07-iteration/iteration-backlog.md (Use template: .agents/templates/planning/iteration-backlog-template.md)

Step 3: Design + Architecture Assessment (parallelizable)

Steps 3a and 3b can run in parallel.

Step 3a: Design Iteration ⟨parallel⟩

Step file: steps/step-03a-design-iteration.md Invoke @product-designer to design improvements:

  • Update user flows based on identified pain points
  • Design A/B test variants (if applicable)
  • Update interaction specs for modified features
  • Document design rationale linking back to data signals

Input: artifacts/output/07-iteration/iteration-backlog.md Output: artifacts/output/07-iteration/iteration-spec.md (Use template: .agents/templates/product/product-spec-template.md — see ## Cross-References section for expected structure)

Step 3b: Architecture Impact Assessment ⟨parallel⟩

Step file: steps/step-03b-architecture-impact.md Invoke @architect to assess:

  • Does the iteration require architectural changes?
  • Are there data model changes needed?
  • Is the change additive or does it modify existing behavior?
  • Flag any technical debt that should be addressed alongside

Input: artifacts/output/07-iteration/iteration-backlog.md Output: artifacts/output/07-iteration/iteration-adr.md (only if architectural changes needed. Use template: .agents/templates/architecture/adr-template.md)

Step 4: Plan and Execute

Step file: steps/step-04-plan-execute.md

Invoke @tech-lead to:

  • Break iteration items into tasks (typically smaller than new features)
  • Estimate effort based on existing velocity
  • Sequence for fastest delivery (quick wins first)

Invoke @product-manager to:

  • Fit iteration work into existing cadence
  • Manage scope and timeline trade-offs
  • Coordinate with any ongoing development work

Then execute the same dev loop from develop:

  • @developer implements
  • @code-reviewer reviews (max 2 review cycles)
  • @qa-engineer validates (max 2 QA cycles)
  • @product-manager verifies

Output: artifacts/output/07-iteration/iteration-plan.md (Use template: .agents/templates/planning/execution-plan-template.md)

Step 5: Ship and Measure

Step file: steps/step-05-ship-measure.md

Invoke @devops-engineer to deploy the iteration. Invoke @data-analyst to measure the impact:

  • Compare metrics before and after iteration
  • Validate that the change achieved its hypothesis
  • Document lessons learned

Write results to memory:

@memory-controller write lessons-learned.md
### [LESSON] {title} [date: YYYY-MM-DD] [agent: @data-analyst]
{what the iteration achieved and what was learned}
**Status:** active

Write session summary:

@memory-controller session-write
Worked on: Iteration cycle — {what was improved}
Decisions made:
- {hypothesis tested}
- {result: positive/neutral/negative}
Next step: {continue iterating / load retro / load develop}
New blockers: {any issues, or "none"}

Output: artifacts/output/07-iteration/iteration-results.md (Use template: .agents/templates/planning/iteration-results-template.md)

Step 6: Decide Next Action (gate)

Step file: steps/step-06-decide-next.md

Based on results:

  • Positive signal: Continue iterating on adjacent improvements → back to Step 1
  • Neutral signal: Revisit the hypothesis, consider larger redesign → back to Step 2
  • Negative signal: Roll back if needed, reanalyze → back to Step 1 with new data
  • Feature stabilized: Move on to next feature or load retro

Iteration limit: Maximum 5 iteration cycles before a mandatory retro to assess whether the iteration strategy is working. This prevents endless polish on a feature that may need a fundamentally different approach. The count lives in artifacts/output/07-iteration/iteration-state.md and is checked by cycle_state_check.js --kind iterate — see Loop closure below.

Output artifacts

  • artifacts/output/07-iteration/analytics-insights.md
  • artifacts/output/07-iteration/iteration-backlog.md
  • artifacts/output/07-iteration/iteration-spec.md
  • artifacts/output/07-iteration/iteration-plan.md
  • artifacts/output/07-iteration/iteration-results.md

When to use

Use this when:

  1. A feature has launched and you have real user data
  2. You need to improve an existing feature based on analytics
  3. You're running A/B tests and want to iterate on winners
  4. You're addressing post-launch user feedback

Key principles

  • Data-driven decisions. Every iteration hypothesis must be measurable.
  • Small batches. Ship small improvements frequently rather than large overhauls.
  • Preserve backwards compatibility. Iterations modify live features — don't break existing users.
  • Close the loop. Always measure the impact of what you shipped before moving on.

Handoff

  • Need to ship a completely new feature? → load develop
  • Production incident? → load incident
  • Want to review how the team is working? → load retro
  • Hit 5 iteration cycles? → mandatory retro before continuing. The count is in artifacts/output/07-iteration/iteration-state.md and cycle_state_check.js --kind iterate blocks a sixth cycle until lastRetroCycle is recorded

Loop closure — checked mechanically

The 5-cycle limit is stated in four places across two skills and was, until now, enforced in none of them: there was no counter, no state file, and nothing that read a number. A limit with no mechanism is not a limit, it is a sentence.

Copy templates/iteration-state.md to artifacts/output/07-iteration/iteration-state.md on the first cycle and increment it as cycles complete. Then:

node .agents/scripts/cycle_state_check.js check --kind iterate \
  --state artifacts/output/07-iteration/iteration-state.md --root . --json

It requires:

  • the counter to agree with the cycle table beneath it — a counter that drifts from its own history is one nobody can audit
  • every counted cycle to carry a positive / neutral / negative result. Counting attempts is how a limit stops limiting, so a cycle with no recorded signal does not count
  • iteration-results.md to record a before/after measurement. The Done-when above already said an unmeasured iteration is not shipped, it is guessed — this checks it
  • at 5 cycles, lastRetroCycle must be recorded before a sixth can start. retro is mandatory at 5, not optional and not deferrable

What it does not check: whether five was the right number, or whether the hypotheses were sound. Those need a human who was there.

Exit 0 conformant · 1 failures · 2 usage/IO error, fail-closed.

State Machine Integration

The pipeline state machine (node .agents/scripts/orchestrator_state.js) is the canonical record of project state. This skill must wire its work into it so other skills, the dashboard, and the code-graph see what happened.

At Start

Run:

node .agents/scripts/orchestrator_state.js status

The project should be in the development phase (or post-launch, depending on how your project is configured). If pipeline is uninitialized, run init first.

At End — Record Completion

Record each iteration artifact:

node .agents/scripts/orchestrator_state.js complete --agent data-analyst --artifact 07-iteration/analytics-insights.md
node .agents/scripts/orchestrator_state.js complete --agent product-manager --artifact 07-iteration/iteration-backlog.md
node .agents/scripts/orchestrator_state.js complete --agent product-designer --artifact 07-iteration/iteration-spec.md
node .agents/scripts/orchestrator_state.js complete --agent architect --artifact 07-iteration/iteration-adr.md
node .agents/scripts/orchestrator_state.js complete --agent tech-lead --artifact 07-iteration/iteration-plan.md
node .agents/scripts/orchestrator_state.js complete --agent developer --artifact 07-iteration/iteration-results.md

Record only the artifacts actually produced. The iteration-results.md triggers an automatic code-graph refresh (because the producing agent is developer), so the next iteration's planning has a current dependency view.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.