Skip to content

reticlehq/reticle / skills/replay-user-flows

replay-user-flows

Verified

Turn a user journey you just clicked through into a saved regression check that re-runs deterministically, with no model in the loop and no test code to write. Use when you have driven the same flow twice, when the user wants regression coverage without a Playwright suite, when a refactor needs proving against every existing journey, or when re-verifying by hand is costing a full drive every time.

Record a journey once, re-verify it forever

Exploring an app to find a journey is the expensive part, and re-driving it with a model pays that cost again on every change. Reticle flows pay it once: the journey is saved with semantic anchors and replayed deterministically afterwards.

Needs Reticle wired in the project. Not set up? Follow the reticle skill: it installs Reticle, asks the user before init changes the app's files, and proves one flow.

Record

You do not have to ask. A drive is saved as a flow automatically when the session ends, written to .reticle/flows/. Commit it: any agent on the repo can then replay it.

What decides whether it is worth committing is how you drove it. A step keeps a consequence only when you declared one, so drive the golden path like this:

reticle_act_and_wait({ sessionId, ref, action: "click", until: { kind: "signal", name: "task:created" } })

That step replays as a test. A bare reticle_act saves a click with nothing to prove, and replays green through any regression.

To name a flow deliberately rather than take the automatic one, reticle_record and reticle_flow_save do it through reticle_run { tool, args }. Neither is advertised, and neither needs to be: reticle_run is on the default surface and dispatches to any registered tool by name. RETICLE_ADVERTISE_ALL_TOOLS=1 advertises them outright, under older, unmerged tool names. Do not set it for an agent: the calls in this skill use the default names.

Annotate the business outcome, not just the clicks, so a replay proves the journey achieved something. Extended surface, like the two above:

reticle_run({ tool: "reticle_annotate", sessionId, args: { flow: "create-task", kind: "intent", text: "create a task and see it in the list" } })
reticle_run({ tool: "reticle_annotate", sessionId, args: { flow: "create-task", kind: "success-state", signal: "task:created" } })

You do not need to add data-testid first. A step whose element has no testid is anchored on its component and source location automatically, and a testid-preserving refactor still replays green.

Replay

reticle_verify({ sessionId, action: "change", files: ["src/tasks/TaskList.tsx"] })

That replays the flows covering those files. To replay one named flow directly, reticle_flow_replay is reached through reticle_run.

Three statuses, and the failures are legible rather than blind:

statusmeansnext
okevery anchor resolved, every expectation helddone
driftan anchor missed: a renamed testid, a signal that never firedread decision.nextAction; it names the file:line and the closest surviving anchor
errorthe flow file is missing or invalid, or a step failed at runtimefix from the error envelope's failed step

On drift, reticle_verify { action: "heal" } proposes the nearest-match rebind so flows do not rot. Apply it when the rename was intentional; treat it as a finding when it was not.

Re-verify the whole suite after any change

reticle_verify({ sessionId, action: "flows" })
// → { status, total, passed, failed, failures: [{ flow, verdict, whatChanged, whereInSource, nextAction }] }

One call, every saved flow, no model per flow. Only failures carry detail, so a green suite is cheap to check. Build → reticle_verify { action: "flows" } → fix from each nextAction → repeat is the regression loop, and it is the point of recording in the first place.

Only the flows your change could have broken

On a large suite, replaying everything after a one-file edit is waste. Hand it the diff instead:

reticle_verify({ sessionId, action: "change", since: "HEAD~1" })

It works out which saved flows cover the files you edited and replays only those. Give it a git ref or the file list. Use this in the inner loop and reticle_verify { action: "flows" } before you ship: the narrow one is fast, the whole one is the guarantee.

Which of your flows actually prove anything

reticle_run({ tool: "reticle_domain", sessionId })   // not advertised, one hop away
// → { flowCount, coverage: { asserted, presenceOnly, assertionFree }, gaps: { declaredUntestedSignals, … } }

A recorded flow that asserts nothing replays green through any regression: it proves the clicks still resolve, not that the app still works. Check this after a recording session: anything landing in assertionFree needs an annotate pass with a success-state, or it is decoration.

When NOT to record

A journey you will run once is cheaper to drive with reticle_act_and_wait and forget. Record the flows that define the product (the ones a regression in would be a bad day) and leave exploratory drives unsaved. A suite of forty half-meant flows costs more attention than it returns.

Honesty

A replay reports what happened. drift is not a pass, and healing a flow to make it green when the app genuinely broke is the one thing that makes the whole suite worthless. If the rename was not intentional, the drift is the finding: report it with the whereInSource pointer.


Full flow reference, one page: curl https://docs.reticle.sh/flows.md. Index of everything: curl https://docs.reticle.sh/llms.txt.

finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall