Index of every page on this site: llms.txt
All skills
launchdarkly avatar

/should-flag-change

@f0ee110 official
Run it yourself

Decide whether a committed code change should ship behind a LaunchDarkly feature flag. Use when a developer asks whether a change should be behind a flag, when reviewing a diff or pull request, or when running in CI on a PR. Runs a fixed decision tree over the change and the code around it, with checked evidence, and emits a structured advisory recommendation. Run it on every change, including test-only, docs, dependency and already-flagged ones: those finish in a minute or two, and its verdict, not your own read of the diff, is the decision. Read-only: it never creates or modifies flags.

Use this Skill: https://skilld.dev/gh/launchdarkly/ai-tooling/should-flag-change

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesRISK.md

≈1.6k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Risk search

You are checking one pull request for concrete risk that a feature flag would let the team turn off: a change to behavior customers already depend on, on a path they reach. Not a judgment that the change "seems risky": a finding you can show, line by line, in the code at the pull request's commit.

THE CHANGE

  • Repository: {CHECKOUT}. Read files at the change's commit only: git -C {CHECKOUT} show {HEAD}:<path>; search with git -C {CHECKOUT} grep -n <pattern> {HEAD} -- <paths>. The working tree may hold other work: don't read it.
  • The pull request's commit: {HEAD}. The version on main it is compared with: {BASE}. See the change with git -C {CHECKOUT} diff {BASE}...{HEAD}.
  • The evidence, including the PR text: {RUN}/evidence.txt (its CHANGED FILES section gives each file's role: shipping code, test, build, docs).

WHAT COUNTS (flag changes to paths already in use, new calls and writes, backend performance changes, fixes that touch stored data or an external integration, a new control that adds a new action) Report a finding only if it is one of these kinds, in SHIPPING code:

  • new-call: a new call or write to another service, external integration, queue or durable store, on a path customers' requests or jobs already run;
  • retarget: an existing call, read or write now sent to a different target (another project, account, deployment, client, endpoint, table or column);
  • cache: a cache placed in front of an existing read on such a path;
  • fail-closed: a new condition that stops, blocks or skips that path when such a call fails or comes back empty;
  • new-action: a new control or entry on an existing customer screen that triggers a new request, write or action (not only local UI state: opening a panel, sorting, collapsing, styling is not an action);
  • removed: a control, option, endpoint or behavior customers use, removed with no replacement where they used it;
  • stored-data: a change to how existing customer data is written, migrated or backfilled.

A NEW FAILURE, NOT JUST A PATH Say for each finding whether the change is meant to keep behavior the same (a refactor, a swapped data source, a reworked query, a moved component): "preserves_behavior": true or false. When it is true, the finding must also name the new failure: the case -- an input, a state, a dependency that errors or comes back empty, a page or a project shape -- in which customers end up worse off at {HEAD} than they did at {BASE}: what the version on main did in that case, and what the change does instead. Such a change is a finding only when the code shows such a case. A change that means to alter behavior (a new call or write, a stored-data change, a new action, a removal) is judged by the kinds above as before. A failure the old code already had in the same case (the same error, empty state or fallback when a read fails) is not new. A bug that merely could exist is not concrete: name the case and the line that produces the worse outcome, or don't report it.

DO NOT REPORT

  • tests, stories, fixtures, docs, build or CI configuration;
  • internal tools, admin consoles and developer utilities that only employees use (an internal launcher, a support console, a private developer CLI);
  • logging, metrics, traces or usage-analytics events only. NOT this exclusion: data that decides how a signup or account is treated downstream -- a classification verdict (for example junk), lead routing, or targeting data sent to sales or marketing systems (Segment traits, Marketo leads) or stored on the account. Report that as stored-data or new-call;
  • a migration or schema change no shipping code reads or writes yet, or code that nothing shipping calls yet (a later pull request will);
  • a change that takes effect when it is deployed or applied, not when code runs: a schema or DDL migration, a table swap or rename, a backfill, build or infrastructure configuration. A runtime flag can't switch it off, so it isn't a finding even when shipping code reads the result. NOT this exclusion: application code in the same change that starts reading or writing somewhere new -- report that part;
  • a change that only affects inputs that already failed on main (a repair of an error path);
  • copy, styling, layout or ordering changes on their own;
  • a change already behind a LaunchDarkly flag the evidence shows enclosing it. If you are unsure whether a path is customer-facing, it isn't a finding.

WHAT TO REPORT For each finding, a chain of real lines at {HEAD}, quoted exactly:

  • at least one CHANGED line: a line the diff adds or modifies, in shipping code (role "changed");
  • the ENTRY a customer reaches it through: a route or page, a request handler bound to a user session or API token, a webhook or job that runs for customers (role "entry"), with any step between them that isn't obvious (a call, a render, a registration; role "calls" / "renders" / "registers"). Every link is checked mechanically: the changed line must be in the diff, the entry must exist at {HEAD} outside tests, and a link that doesn't match drops the finding. An empty list is the right answer when nothing qualifies: most changes have none. Don't stretch a kind to fit.

Read-only: don't modify the repository, create or change flags, or use the network.

Write ONLY this JSON to {RUN}/risks.json, then run python3 {SKILL_DIR}/ld-factory.pyz risks {RUN}:

{"findings": [{"kind": "new-call | retarget | cache | fail-closed | new-action | removed | stored-data", "chain": [{"path": "path/at/head", "line": 123, "text": "the exact line", "role": "changed | entry | calls | renders | registers"}], "preserves_behavior": true, "new_failure": {"case": "the input, state or failure in which it differs", "main": "what customers got at the version on main", "head": "what they get with the change"}, "customer_facing_because": "who reaches the entry, and how (a session-bound route, a customer page, a job per customer account)", "why": "one sentence, in plain words about the developer's code: what could go wrong for customers that a flag would let the team switch off"}], "searched": "one or two sentences: what you checked"}

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 alerttoday3 checks · Risk MEDIUM
  • Gen Agent Trust Hubtoday

    This skill utilizes a bundled executable that contains hidden Unicode characters, which could be used for malicious steganography, and it analyzes untrusted data from repositories and pull requests, making it susceptible to indirect prompt injection.

  • Sockettoday

    1 alert: gptSecurity

  • Snyktoday

    Risk: CRITICAL · 2 issues

Signed by skilld at f0ee110. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 days ago
metadata
{
  "author": "launchdarkly",
  "version": "0.5.3",
  "graph": "07355490e985"
}
Other metadata
compatibility
Advisory and read-only. Needs a git checkout with the change committed (shallow is fine; a --filter partial clone is filled in from its remote), a shell, git and Python 3.10+ (no packages). Reads LaunchDarkly flag state through your LaunchDarkly tools when available, and works without them. When it can't run (no shell, no checkout), it decides by judgment instead, from references/NO_CHECKOUT.md. Returns a structured verdict as a `recommend-flag` block.

README badge

README badge for launchdarkly/ai-tooling/should-flag-change