agentleFS
PackmindHubPackmindHub/packmind/.cursor/skills/feature-flags-audit/SKILL.md

feature-flags-audit

Skill321 starsChanged 54 days agoApache-2.0View on GitHub

Audit and inventory all feature flags declared in the Packmind codebase. Use when the user asks to list, audit, review, or inventory feature flags, asks which flags are active, wants to know what a flag gates, or asks which flags are opened to a given domain/user. Produces a synthetic markdown table with each flag key, its audience, an inferred functional description, its active/orphan status, and its usage locations.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 3 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If a flag is not declared in DEFAULT_FEATURE_DOMAIN_MAP, it does not exist.

    line 25
  2. Very specific:

    Every exported constant whose name ends with _FEATURE_KEY, along with its string value.

    line 33
  3. Very specific:

    Cross-check constants against the map: Every _FEATURE_KEY constant should have a matching entry in DEFAULT_FEATURE_DOMAIN_MAP, and every …

    line 38
  4. Distinctive:

    If you find either case, do not drop it from the inventory — include it and surface the mismatch in the Summary section.

    line 42
  5. Distinctive:

    Write one short sentence in the functional voice — describe what the user sees (or doesn't see) when the flag is off.

    line 76
  6. Very specific:

    If you cannot form a confident one-liner after reading the usage sites, write a shorter description prefixed with (uncertain) so the user …

    line 87
  7. Very specific:

    Add an E2E tested column to the output table (Step 5) with either Yes or No.

    line 99
  8. Distinctive:

    After the table, add a short Summary section with:

    line 132
  9. Very specific:

    Do not invent flags: Only entries in DEFAULT_FEATURE_DOMAIN_MAP are real flags.

    line 154
  10. Very specific:

    Audience targeting is email-only: Do not describe the audience as "organizations", "plans", "teams", or "percentages" — those concepts do …

    line 155

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

Feature Flags Audit

Produce a synthetic inventory of every feature flag declared in the Packmind codebase, with audience targeting, inferred functional descriptions, and orphan detection. The result is a single markdown table the user can scan at a glance.

Context

Packmind uses a custom feature flag system shared by both frontend and backend. There is no third-party library (LaunchDarkly, GrowthBook, PostHog, etc.). Flags are gated by email domain or exact email via a single central registry. There is no org-based, plan-based, or percentage-rollout mechanism — audience targeting is email-based only. One runtime override does exist: an FF_* env kill-switch (see below) can force a flag fully on or off regardless of the email rule.

Canonical source of truth: packages/feature-flags/src/registry.ts (the shared, browser-safe @packmind/feature-flags package)

This file exports:

  • Flag key constants ending in _FEATURE_KEY (e.g. SPACES_MANAGEMENT_FEATURE_KEY = 'spaces-management').
  • The FeatureFlagKey union of all valid flag string values.
  • DEFAULT_FEATURE_DOMAIN_MAP: a Record<string, readonly string[]> mapping each flag's string value to a list of allowed entries. Entries starting with @ are domain rules (e.g. @packmind.com matches any user at that domain). Entries containing @ but not starting with @ are exact email matches.
  • The isFeatureFlagEnabled() pure function used at call sites (defined in packages/feature-flags/src/isFeatureFlagEnabled.ts).

The registry is consumed on both sides: the frontend imports it via @packmind/ui (which re-exports the keys, DEFAULT_FEATURE_DOMAIN_MAP, and isFeatureFlagEnabled for back-compat; packages/ui/src/lib/components/content/PMFeatureFlag/PMFeatureFlag.tsx now holds only the PMFeatureFlag React wrapper component). The backend evaluates the same registry through the isFeatureEnabled(flag, { userEmail }) helper in @packmind/node-utils (packages/node-utils/src/featureFlags/isFeatureEnabled.ts), which layers the FF_* env kill-switch on top of the shared email-domain rule.

If a flag is not declared in DEFAULT_FEATURE_DOMAIN_MAP, it does not exist. Do not invent flags.

Workflow

Step 1 — Parse the canonical registry

Read packages/feature-flags/src/registry.ts. Extract:

  1. Every exported constant whose name ends with _FEATURE_KEY, along with its string value. The string value is what appears as a key in DEFAULT_FEATURE_DOMAIN_MAP — keep the mapping (constant name ↔ string value) in memory.
  2. The full DEFAULT_FEATURE_DOMAIN_MAP object. For each entry, record the flag's string value and its list of allowed domains/emails.

Build an internal list of flags with three fields so far: constant name, string value, audience (array of allowed entries).

Cross-check constants against the map. Every _FEATURE_KEY constant should have a matching entry in DEFAULT_FEATURE_DOMAIN_MAP, and every map entry should correspond to a declared constant. A mismatch is a real misconfiguration:

  • A constant declared with no map entry produces a flag that is never enabled for anyone — dead code.
  • A map entry whose string value isn't tied to any exported constant means call sites can't import it symbolically and are likely using the inlined string (fragile).

If you find either case, do not drop it from the inventory — include it and surface the mismatch in the Summary section.

Step 2 — Find usages for each flag

For each flag, search the codebase for where it is consumed. Use Grep:

  • First pass: search for the constant name (e.g. SPACES_MANAGEMENT_FEATURE_KEY). This is the canonical import form and catches almost all usages.
  • Second pass: search for the string value (e.g. 'spaces-management' or "spaces-management") as a safety net. Some usages may inline the string instead of importing the constant.
  • Backend pass: search for isFeatureEnabled( (the backend helper call) across apps/api/**, apps/mcp-server/**, apps/cli/**, and packages/**. This is a plain async helper call, not an injected port — so search the function name, not an interface. A call looks like await isFeatureEnabled('some-flag', { userEmail }); the first argument tells you which flag is gated. (Backend call sites may currently be zero — the mechanism exists but the first real consumer may not have landed yet — so finding none is normal, not a bug.)

Scope the search to:

  • apps/frontend/src/**
  • apps/frontend/app/** — the React Router v7 file-based route tree. Route modules live here (e.g. apps/frontend/app/routes/org.$orgSlug._protected.settings.tsx), not under src/, and they gate flags too (typically via isFeatureFlagEnabled(...) to show/hide nav entries, routes, or whole pages). Omitting this directory will mis-report a live route-gated flag as an orphan.
  • apps/api/**, apps/mcp-server/**, apps/cli/** — backend use cases may gate flags via isFeatureEnabled(flag, { userEmail }). Include these so a backend-gated flag is not mis-reported as an orphan.
  • packages/** — includes the shared @packmind/feature-flags package (registry) and @packmind/node-utils (the backend helper); the registry definition files are excluded below.

Exclude:

  • The canonical registry files themselves: packages/feature-flags/src/registry.ts and packages/feature-flags/src/isFeatureFlagEnabled.ts (plus the package barrel packages/feature-flags/src/index.ts, which only re-exports them).
  • The backend helper definition itself: packages/node-utils/src/featureFlags/isFeatureEnabled.ts (it defines isFeatureEnabled; it is not a consumer). Count only call sites of the helper, not its declaration.
  • The @packmind/ui back-compat re-export shim: packages/ui/src/lib/components/content/PMFeatureFlag/PMFeatureFlag.tsx (now only the React wrapper + re-exports) and its co-located test file PMFeatureFlag.test.tsx. (The parent barrel at packages/ui/src/lib/components/content/index.ts re-exports for back-compat but is not a real gate.)
  • Test and spec files anywhere in the search scope: *.test.*, *.spec.*, and Storybook stories *.stories.*. A usage inside a test doesn't represent shipping behavior and would mislead a reader scanning the "Usage files" column. If a flag is only referenced from tests, treat it as an orphan (see Step 4) and note the test-only state in the description.
  • Generated/build artifacts (dist/, node_modules/).

Record the list of files where each flag is actually consumed.

Step 3 — Infer a functional description

For each flag with at least one usage, open 1 to 3 of the most representative usage files and read enough context to understand what the flag actually gates from a user-facing perspective. Use these signals (in rough priority order):

  1. The JSX wrapped inside <PMFeatureFlag featureKeys={[FLAG]}>...</PMFeatureFlag> — this is literally what appears or disappears.
  2. The branch or value guarded by isFeatureFlagEnabled({ featureKeys: [FLAG] }) — often an "edit" / "create" / "delete" affordance, or an entire route/tab.
  3. The component or file name hosting the gated code (e.g. SpacesManagementActions.tsx → space management actions).
  4. Button labels, section titles, or surrounding comments near the gate.

Write one short sentence in the functional voice — describe what the user sees (or doesn't see) when the flag is off. Avoid technical framing.

Examples of good descriptions:

  • "Reveals the 'Move to space' action and space management UI in the recipes list."
  • "Enables inline editing of change proposals in review detail pages."
  • "Adds 'Propose change' links from recipes and standards to the webapp review flow."

Examples of bad descriptions (do not write these):

  • "Wrapped by the PMFeatureFlag component in SpacesManagementActions.tsx." (tells the reader nothing functional)
  • "Feature flag for spaces management." (circular, just re-states the key)

If you cannot form a confident one-liner after reading the usage sites, write a shorter description prefixed with (uncertain) so the user knows to double-check.

Step 3b — Check E2E test coverage

For each flag, search apps/e2e-tests/src/ for spec files that call .use({ underFeatureFlag: true }). These tests explicitly target a feature-flagged flow and represent intentional E2E coverage of the flag.

grep -r "underFeatureFlag: true" apps/e2e-tests/src/

Record which flags have associated E2E test coverage. A flag with no E2E coverage is not an orphan (orphan is determined by frontend usage, not tests), but the information is worth surfacing so the user knows which flagged features are tested end-to-end.

Add an E2E tested column to the output table (Step 5) with either Yes or No.

Step 4 — Detect orphans

A flag is orphan if, after the two-pass search, no usage is found outside the canonical file and its test. Orphans are candidates for removal and worth surfacing — they often accumulate when a feature ships and nobody remembers to delete the gate.

Before declaring a flag orphan, confirm the search covered apps/frontend/app/** (the route tree), not just apps/frontend/src/**. A flag is frequently consumed only in a route module to gate a nav entry or page — missing that directory is the most common cause of a false orphan.

For orphans:

  • Set status to Orphan.
  • Set the description to (no usage found — candidate for removal).
  • Set the usage-files column to —.

Step 5 — Render the table

Output a single markdown table. Use these columns, in this order, with these exact headers:

| Key | Audience | Functional description | Status | E2E tested | Usage files |

Column rules:

  • Key — the string value (e.g. change-proposals), not the constant name. It's what appears in the domain map and what a PM or stakeholder recognises.
  • Audience — the allowed entries, comma-separated, in the same form as in the map (domains keep their @ prefix). If the list is empty, write (none).
  • Functional description — one short sentence, functional voice. See Step 3.
  • Status — either Active or Orphan.
  • E2E tested — Yes if at least one spec file in apps/e2e-tests/src/ uses .use({ underFeatureFlag: true }), otherwise No.
  • Usage files — comma-separated list of repo-relative paths. Truncate to the 5 most representative ones and append (… +N more) if there are more. "Most representative" is a ranked preference, in this order:

    1. Files containing a <PMFeatureFlag featureKeys={[FLAG]}>...</PMFeatureFlag> JSX gate — these are the literal "what appears when the flag is on".
    2. Files calling isFeatureFlagEnabled({ featureKeys: [FLAG] }) in a branch that guards an action, route, or tab.
    3. Files that import the constant but only pass it downstream (e.g. prop-drilling, utility wiring) — lowest priority, include only if the first two tiers don't fill the 5 slots.

    For orphans, write —.

After the table, add a short Summary section with:

  • Total flag count, active count, orphan count.
  • Any audience anomaly worth flagging — e.g. a flag opened to an external company's domain (like @monpetitplacement.fr, @ninaa.io) is usually a pilot and the user will want to notice it. Mention it briefly.

Output format

# Feature Flags Inventory

| Key | Audience | Functional description | Status | E2E tested | Usage files |
| --- | --- | --- | --- | --- | --- |
| space-identity | @packmind.com, @promyze.com | Reveals the space identity settings panel and related configuration affordances. | Active | No | apps/frontend/src/domain/spaces/components/SpaceGeneralSettings.tsx, apps/frontend/src/domain/spaces/components/SpaceIdentityForm.tsx |
| change-proposals | @packmind.com, @promyze.com, @monpetitplacement.fr, @ninaa.io | Enables the change proposal review flow end-to-end (navigation entry points and review detail pages). | Active | Yes | apps/frontend/src/domain/change-proposals/components/StandardReviewDetail/StandardReviewDetail.tsx, apps/frontend/src/domain/change-proposals/components/SkillReviewDetail/SkillReviewDetail.tsx, apps/frontend/src/domain/change-proposals/components/CommandReviewDetail/CommandReviewDetail.tsx (… +2 more) |
| some-old-flag | @packmind.com | (no usage found — candidate for removal) | Orphan | No | — |

**Summary:** 7 flags — 6 active, 1 orphan. `change-proposals` is opened to `@monpetitplacement.fr` and `@ninaa.io` (external pilots, worth monitoring).

Output only the report — no preamble, no "here is the table" sentence. The user invokes this skill to get the report itself.

Notes and edge cases

  • Do not invent flags. Only entries in DEFAULT_FEATURE_DOMAIN_MAP are real flags. If the user asks about a flag that isn't there, say so explicitly.
  • Audience targeting is email-only. Do not describe the audience as "organizations", "plans", "teams", or "percentages" — those concepts do not exist in this system. A domain like @packmind.com means "any user whose email is at this domain", nothing more.
  • The registry is the audience source of truth, but not fully static. There is no remote config or third-party service, but an FF_* env kill-switch does exist: the backend helper reads Configuration.getConfig('FF_' + SCREAMING_SNAKE(flag)) (e.g. FF_CHANGE_PROPOSALS_IN_WEBAPP) — values on/all/true force the flag fully on, off/none/false force it off, and unset/empty falls through to the email-domain rule in DEFAULT_FEATURE_DOMAIN_MAP. So the map defines the default audience; an env override can win over it at runtime.
  • The backend uses feature flags too. Scan apps/api/, apps/mcp-server/, and apps/cli/ for backend call sites of isFeatureEnabled(flag, { userEmail }) (the @packmind/node-utils helper). This is a plain async function call, not an injected port — search for the function name, not an interface. Backend call sites may currently be zero (the mechanism is in place ahead of its first real consumer), so finding none is expected, not a violation.
  • Frontend code is split across two roots. Components and domain logic live in apps/frontend/src/, but React Router v7 route modules live in apps/frontend/app/routes/. Flags are gated in both. Searching only src/ is the classic miss — e.g. MARKETPLACES_FEATURE_KEY is consumed solely in apps/frontend/app/routes/org.$orgSlug._protected.settings.tsx to gate the Settings → Distribution → Marketplaces nav entry, and would look like an orphan if app/ were skipped.
  • PMFeatureFlag with empty featureKeys={[]} always renders. This is a rare pattern but if you see it while reading usages, don't count it as a real gate.

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.