Click Path Audit Edho Ferdian

Click-Path Audit — Edho Ferdian Mode

Ordinary debugging asks: does the handler exist, does it crash, are the types right. All three can be YES while the button is still broken. This skill asks the question those three miss:

> Does the final state match what the control's label promises?

The defect class

A "New Email" button called setComposeMode(true) and then selectThread(null). Both functions existed, neither threw, types were correct — and selectThread reset composeMode: false as an undeclared side effect. The button did nothing. A systematic debugging pass that found 54 other defects missed this one, because nothing about it is visible from either function in isolation. It only appears when you look at the sequence.

Scope first

This audit is expensive; scope it before starting:

| Scope | When | |-------|------| | One control | A user reported one specific broken button | | One screen | A new page was just built, or one screen misbehaves | | One store | A shared store action was modified — audit every caller of the changed actions | | Whole app | Pre-release, or after a refactor that touched shared state broadly |

For a whole-app audit, Step 1 must complete before any Step 2 work begins — its output is the input for everything else. On a harness with sub-agent delegation, build the map yourself, then fan out Step 2 to click-path-tracer-edho-ferdian in parallel — one delegate per screen/module, each given the complete map, never building its own partial one. This is worth the delegation overhead at whole-app scale (potentially dozens of touchpoints, each independent to trace once the map exists) but not for the smaller scopes above — trace those inline. On a harness with no delegation primitive, trace every touchpoint inline regardless of scope, per Step 2 below.

Step 1 — Build the side-effect map (mandatory, always first)

For every state store in scope (Zustand store, Redux slice, React context reducer, signal container, view model), record for each action:

• which fields it sets
• which fields it resets as a side effect — fields it does not conceptually

own

STORE: emailStore
  setComposeMode(bool)  → sets {composeMode}
  selectThread(t|null)  → sets {selectedThread, selectedThreadId, messages, drafts}
                          RESETS {composeMode: false, composeData: null, redraftOpen: false}
  setDraftGenerating(b) → sets {draftGenerating}

DANGEROUS RESETS (an action clearing state it does not own):
  selectThread → resets composeMode (owned by setComposeMode)
  reset        → resets everything

The DANGEROUS RESETS list is the whole point of this step. Every defect this skill exists to catch lives there. Do not proceed to Step 2 with an incomplete map — an audit against a partial map produces false confidence, which is worse than no audit.

Step 2 — Trace each touchpoint

For every interactive element in scope (onClick, onSubmit, onChange, keyboard handler, gesture handler):

TOUCHPOINT: "New Email" — ThreadList.tsx:88
  HANDLER: onClick
    1. setComposeMode(true)   → sets {composeMode: true}
    2. selectThread(null)     → RESETS {composeMode: false}   ← CONFLICT
  EXPECTED (from the label): a blank compose form opens
  ACTUAL: composeMode is false; nothing renders
  VERDICT: BUG — Sequential Undo

Check every trace against these six patterns:

1. Sequential undo — a later call resets what an earlier call set. The

canonical case above.

1. Async race — two async calls both write the same field; the final value

depends on resolution order, not on intent.

1. Stale closure — a memoized handler captures an old value, so two

increments apply the same stale base and the effect happens once, not twice.

1. Missing transition — the handler validates, logs, or sets a flag but

never performs the action the label promises (no API call, no navigation, no persistence).

1. Conditional dead path — the real work sits behind a condition that is

always false at that point in the lifecycle.

1. Effect interference — the handler sets a field and a watcher/effect

observing that field immediately resets it.

For each call in a trace, answer four questions: what does it read, what does it write, does it touch shared state, and does it reset anything as a side effect.

Step 3 — Report

CLICK-PATH-001 [HIGH]
  Touchpoint: "New Email" — src/components/ThreadList.tsx:88
  Pattern:    Sequential Undo
  Trace:
    1. setComposeMode(true) → sets {composeMode: true}
    2. selectThread(null)   → RESETS {composeMode: false}
  Expected: blank compose form opens
  Actual:   nothing renders; composeMode is false by the end of the handler
  Fix:      reorder (selectThread first, then setComposeMode), or remove the
            undeclared composeMode reset from selectThread and clear it at the
            call sites that actually mean to

Severity: CRITICAL — a money, auth, or destructive control that silently does nothing (the user believes the action happened). HIGH — a primary control that does nothing. MEDIUM — a control whose final state is partly wrong. LOW — a redundant call with no user-visible effect (dead code, not a defect).

Boundaries — what this is not for

• API-level defects (wrong response shape, missing endpoint) — a click-path

trace ends at the request; take those to normal debugging, and to test-authoring-edho-ferdian/references/regression-testing.md for the test.

• Styling and layout — visual inspection, or a QA sweep

(e2e-testing-edho-ferdian/references/qa-sweep.md).

• Performance — performance-audit-edho-ferdian.

Handoff — every finding earns a test

A click-path finding is a proven defect with a known reproduction, which makes it the ideal RED gate. Hand each one to dev-kickoff-edho-ferdian's normal PLAN → TEST → IMPLEMENT loop: the failing test asserts the final state after the handler runs, not that each individual function was called. A test that asserts setComposeMode was called would have passed against the original bug.

Language routing (fixed — see skill-authoring-edho-ferdian's canonical contract)

Communication to the user in Bahasa Indonesia; the trace report and call sequence in English, since it gets pasted into an issue or handed to dev-kickoff-edho-ferdian's TEST/IMPLEMENT loop. Full contract: skill-authoring-edho-ferdian §7.

Global rules

1. Store map before traces, always. No exceptions; a partial map hides the

exact defect this skill targets.

1. Trace in execution order. The order is the evidence.
2. Judge against the label, not against the code. The label is the

contract with the user.

1. A finding is a trace, not an opinion. Every report shows the numbered

call sequence with the conflicting write marked.

1. Do not fix inside the audit. Report, then hand off — an audit that

starts editing loses its own coverage.

Growth path

This skill is intentionally single-file. If the defect-pattern list (Step 2) grows past six patterns, or this file passes ~250-300 lines, split framework- specific tracing detail (e.g. a Zustand/Redux-specific vs. a signals-specific lens) into references/, following the domain+lens pattern other skills in this ecosystem use — do not let a single growing file replace that split.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: edhoferdian
• Source: edhoferdian/EEF
• License: MIT

Install and usage instructions live in the source repository linked above.