SKILL.md
Decide if a note is needed
Required: breaking API change, new feature, bug fix, deprecation, dependency upgrade — if it's customer-impacting. Not required: CI chores, internal-only/not-yet-released API changes, test-only changes.
If customer impact is unclear, ask the user what the customer impact is. Never guess — unless the PR/issue/ticket you were given already states it. If no note is needed, say so and suggest the changelog/no-changelog PR label instead.
Audience
The note is for a customer upgrading dd-trace-py — never for a dd-trace-py contributor. It is not a PR summary. Internal rationale, implementation detail, and CI/test notes go in the PR description, not here.
Before writing
- If the category (
features/fixes/upgrade/deprecations/issues/api/security/other)
isn't given, ask.
- Check for an existing fragment on the same topic, scoped to your current branch's own
unmerged changes: git diff --name-only <base>...HEAD -- releasenotes/notes/, where <base> is the branch you're merging into (usually origin/main; use the actual base if this branch forked off a release branch instead). Only reuse a fragment from that diff — one already on the base branch may already be released; create a new fragment instead of editing a shipped one.
- Read the PR/commit diff for ground truth, but translate it — don't copy engineering language.
Write the note
Generate the skeleton: riot run reno new <slug> (slug: lowercase, hyphenated, e.g. fix-aioredis-catch-canceled-error). Edit the generated YAML, keeping only the section(s) that apply — one Reno fragment per change.
Format: <scope>: <sentence(s)>. Scope is the component name (see :ref:releasenotesscope in docs/releasenotes.rst — tracing, profiling, ASM, dynamic instrumentation, CI visibility, or the integration name). Use internal when the change genuinely isn't tied to one product (e.g. core threading/fork-safety affecting several products) — that's a legitimate scope, not a placeholder for when you didn't pick one.
Fix — state the customer-visible symptom, not the root cause, in present tense: Fixes an issue where <symptom> occurs when <condition>.
Feature/behavior change — if it's configurable or changes a default, state the exact env var / config option / argument and what it does. Don't skip this when a toggle exists.
Never include: internal function/class names, internal tag prefixes (_dd.*), RFC/JIRA/ticket numbers, CI job names, draft leftovers (TODO, XXX, unfinished sentences), or "why we built it this way" rationale.
Length: 1-2 sentences. Longer only to show a code example for a new public API.
Verify facts against the code, not the PR title — defaults and numbers drift.
Examples
Good (fix):
fixes:
- |
profiling: Fixes inconsistent handling of ``DD_PROFILING_MAX_FRAMES`` by clamping it
before stack collection so samples stay within the backend's 600-location limit.
Good (feature, toggle):
features:
- |
AI Guard: Adds per-integration kill switches ``DD_AI_GUARD_ANTHROPIC_ENABLED`` and
``DD_AI_GUARD_LANGCHAIN_ENABLED`` (default ``true``). Set to ``false`` to disable AI
Guard for that provider/framework.
Bad → good rewrite (internal scope is fine here — not product-specific; the fix is the wording):
# Bad: internal mechanism, no customer symptom
fixes:
- |
internal: Fixed a GC reference-cycle leak in PeriodicThread.
# Good
fixes:
- |
internal: Fixes a memory leak that causes memory usage to grow over time in
long-running processes using background periodic tasks.