update-skia

name

update-skia

description

Update the Skia graphics library to a new Chrome milestone in SkiaSharp's mono/skia fork. Handles upstream merge, C API shim fixes, binding regeneration, C# wrapper updates, and dual-repo PR coordination. Use when user asks to: - Update/bump Skia to a new milestone (m120, m121, etc.) - Merge upstream Skia changes - Update the Skia submodule to a newer version - Check what Skia milestone is current or what version of Skia is used Triggers: "update skia", "bump skia", "skia milestone", "update to m121", "merge upstream skia", "skia update", "new skia version", "what milestone", "what version of skia", "current skia version", "check skia version". For updating individual dependencies (libpng, zlib, etc.), use `native-dependency-update` instead. For security audits, use `security-audit` instead.

Update Skia Milestone Skill

Update Google Skia to a new Chrome milestone in SkiaSharp's mono/skia fork.

Key References

• references/breaking-changes-checklist.md — How to analyze breaking changes between milestones
• references/known-gotchas.md — Hard-won lessons and troubleshooting table
• references/typical-changes.md — Files typically changed during an update
• documentation/dev/dependencies.md — Dependency tracking and cgmanifest.json format
• RELEASE_NOTES.md in upstream Skia — Official Skia release notes

Scripts

• scripts/update-versions.ps1 — Phase 6: Updates all version files and runs verification
• scripts/regenerate-bindings.ps1 — Phase 8: Regenerates bindings, reverts HarfBuzz, reports new functions

Overview

Updating Skia is the highest-risk operation in SkiaSharp. It touches:

• The native C++ Skia library (upstream merge)
• The custom C API shim layer (must be adapted to new C++ APIs)
• Generated P/Invoke bindings
• C# wrapper code
• All platforms (macOS, Windows, Linux, iOS, Android, etc.)

Go slow. Research first. Build and test before any PR.

⚠️ Follow Every Phase In Order

This is an 11-phase workflow where each phase builds on the previous one. The phases exist because Skia updates touch four layers (C++ → C API → generated bindings → C# wrappers) and two repositories (mono/skia + mono/SkiaSharp). Skipping a phase doesn't just risk a build failure — it risks shipping broken binaries to customers who won't see the problem until runtime.

Each phase ends with a gate — a verification step that confirms the phase completed correctly. Re-read each phase's instructions before executing it, because the details are project-specific and easy to get wrong from memory.

What You're About to Do

A. Research (Phases 1–3)
   1. Discovery & Current State
   2. Breaking Change Analysis
   3. Validation

B. Branch & Merge (Phases 4–5)
   4. Branch Setup
   5. Upstream Merge

C. Update & Build (Phases 6–7)
   6. Update Version Files
   7. Fix C API Shim & Build Native

D. Regenerate & Verify (Phases 8–10)
   8. Regenerate Bindings
   9. Fix C# Wrappers
  10. Build & Test

E. Ship (Phase 11)
  11. Create PRs

Critical Rules

🛑 STOP AND ASK before: Merging PRs, Force pushing, Deleting branches, Any destructive git operations

🚫 BRANCH PROTECTION (MANDATORY)

Repository	Protected Branches	Action Required
mono/SkiaSharp (parent)	main	Create feature branch first
mono/skia (submodule)	main, skiasharp	Create feature branch first

❌ NEVER Do These

Shortcut	Why It's Wrong
Push directly to skiasharp or main	Bypasses PR review and CI
Skip breaking change analysis	Causes runtime crashes for customers
Use externals-download after C API changes	Causes EntryPointNotFoundException
Merge both PRs without updating submodule in between	Squash-merge orphans commits
Skip tests	Untested code = broken customers

---

A. Research (Phases 1–3)

Phase 1: Discovery & Current State

1. Identify current milestone:

   grep SK_MILESTONE externals/skia/include/core/SkMilestone.h
   grep "^libSkiaSharp.*milestone" scripts/VERSIONS.txt
   grep chrome_milestone cgmanifest.json
   

2. Identify target milestone from user request.

3. Check for existing PRs — Search both mono/SkiaSharp and mono/skia for open update PRs.

4. Verify upstream branches exist and fetch:

   cd externals/skia
   git remote add upstream https://github.com/google/skia.git 2>/dev/null
   git fetch upstream {UPSTREAM_REF}   # chrome/m{TARGET} for a milestone, or main for tip mode
   

   {UPSTREAM_REF} is the upstream ref you merge from: chrome/m{TARGET} for a normal milestone/release-line update, or main for a main/tip sync (see step 5).

   Note: When this phase is pre-computed by the automated workflow, you still need to add the upstream remote and fetch {UPSTREAM_REF} — Phase 5 depends on it.

5. Determine the base branch (main vs a release line). Most updates target main (newest, in-development line) with skiasharp as the mono/skia base. But when the target milestone is older than main's milestone AND a matching release branch exists, the update targets that release line instead:

   # SkiaSharp uses release/<major>.<milestone>.x (e.g. release/4.148.x for m148)
   git ls-remote --heads origin "refs/heads/release/*.{TARGET}.x"
   # mono/skia mirrors the SAME branch name
   git -C externals/skia ls-remote --heads origin "refs/heads/release/*.{TARGET}.x"
   

   Variable	main target	release-line target	main/tip mode
   {UPSTREAM_REF} (merge from)	chrome/m{TARGET}	chrome/m{TARGET}	main (HEAD)
   {BASE_BRANCH} (SkiaSharp)	main	release/<major>.{TARGET}.x	main
   {SKIA_BASE_BRANCH} (mono/skia)	skiasharp	release/<major>.{TARGET}.x	skiasharp
   {HEAD_BRANCH} (both repos)	skia-sync/m{TARGET}	skia-sync/release-<major>.{TARGET}.x	skia-sync/main

   The table's three columns are the three possible sync modes. main target (the default) is a normal milestone bump; the other two are special cases:

   Release-line target — chosen when {TARGET} is older than main's milestone AND a matching release/<major>.{TARGET}.x branch exists. Always a bug-fix-only sync (CURRENT == TARGET): do NOT bump the milestone/soname/nuget version in Phase 6 — only cgmanifest.json's commit hash changes.

   🛑 The release branch must already exist in BOTH repos. Branch creation is owned by the release process (release-branch skill), not this skill. If the SkiaSharp release/<major>.{TARGET}.x branch exists but the mono/skia one does NOT, STOP and fail — do not create it here. Ask a human to cut the mono/skia release branch first.

   main (tip) mode — instead of a chrome/m{TARGET} milestone branch, merge the very tip of upstream google/skia main (HEAD). Like a main target it uses {BASE_BRANCH} = main and {SKIA_BASE_BRANCH} = skiasharp, but with {HEAD_BRANCH} = skia-sync/main and {UPSTREAM_REF} = main, and CURRENT == TARGET, so it is NOT a version bump — keep the milestone/soname/nuget versions unchanged. Unlike a release-line bug-fix sync it MAY still include new upstream API/binding changes, so regenerate + build + test as normal (Phases 8–10). A tip merge is large and conflict-heavy because the submodule base is well behind main, so the verify-upstream-or-reapply policy (Phase 5 / gotcha #15) is mandatory. Because the tip is not a milestone branch, the milestone-pair diff in Phase 2 step 4 doesn't apply — substitute $(git merge-base {SKIA_BASE_BRANCH} upstream/main)..upstream/main (Phase 5's diffs already use {UPSTREAM_REF}, so they need no change).

   Use these {UPSTREAM_REF} / {BASE_BRANCH} / {SKIA_BASE_BRANCH} / {HEAD_BRANCH} values everywhere below in place of the hardcoded chrome/m{TARGET} / main / skiasharp / skia-sync/m{TARGET} defaults.

🛑 GATE: Confirm current milestone, target milestone, the base branch (main vs release line, with the mono/skia base branch confirmed to exist), and that the upstream branch exists.

Phase 2: Breaking Change Analysis

This is the most critical phase. Thorough analysis here prevents customer-facing breakage.

1. Read official release notes for EVERY milestone being skipped:

   • Fetch https://raw.githubusercontent.com/google/skia/main/RELEASE_NOTES.md
   • Document all changes for each milestone between current and target

2. Categorize changes by impact:

   Category	Risk	Examples
   Removed APIs	🔴 HIGH	Functions deleted, enums removed
   Renamed/Moved APIs	🟡 MEDIUM	Namespace changes, header moves
   New APIs	🟢 LOW	Additive changes, new factories
   Behavior changes	🟡 MEDIUM	Default changes, semantic shifts
   Graphite-only	⚪ SKIP	SkiaSharp uses Ganesh, not Graphite

3. Map each HIGH/MEDIUM change to C API files:

   cd externals/skia
   # Check which C API files reference affected APIs
   grep -r "GrMipmapped\|GrMipMapped" src/c/ include/c/
   grep -r "refTypefaceOrDefault\|getTypefaceOrDefault" src/c/ include/c/
   

4. Run structural diff on include/ directory:

   git diff upstream/chrome/m{CURRENT}..upstream/chrome/m{TARGET} --stat -- include/
   git diff upstream/chrome/m{CURRENT}..upstream/chrome/m{TARGET} -- include/core/ include/gpu/ganesh/
   

👉 See references/breaking-changes-checklist.md for the full analysis template, including verification steps for struct sizes, moved files, and diff-reading traps.

🛑 GATE: Include the breaking change analysis in the PR description body. Summarize the key findings (HIGH/MEDIUM risk changes and their C API impact) for the user.

Phase 3: Validation

The agent performing the breaking change analysis has blind spots — it may filter out relevant changes or miss moved headers. An independent validation catches these before they become runtime crashes.

Launch an explore agent with model: "claude-opus-4.6" using the prompt template from references/validation-prompt.md — substitute the milestone numbers and paste your breaking change analysis table. The default explore model (Haiku) is too weak for accurate header-level validation — use Opus for reliability.

🛑 GATE: Validation agent has run and confirmed analysis. If it found missed items, update the analysis and re-present to user before proceeding.

✅ Before proceeding to B (Branch & Merge):

• Current and target milestones confirmed
• Breaking change analysis complete
• Validation passed

---

B. Branch & Merge (Phases 4–5)

Phase 4: Branch Setup

⚠️ This phase creates BOTH branches before making ANY changes. You may be on a workflow branch, a feature branch, or a detached HEAD — none of which is the right base. You MUST branch from origin/{BASE_BRANCH} (parent) and origin/{SKIA_BASE_BRANCH} (submodule). For a main update (and main/tip mode) those are origin/main and origin/skiasharp; for a release-line update they are origin/release/<major>.{TARGET}.x in both repos (see Phase 1 step 5).

Parent repo (SkiaSharp):

1. Fetch the latest base branch:

   git fetch origin {BASE_BRANCH}
   

2. Create the feature branch from origin/{BASE_BRANCH}:

   git checkout -b {HEAD_BRANCH} origin/{BASE_BRANCH}
   

   If the branch already exists on the remote, check it out instead:

   git fetch origin {HEAD_BRANCH} && git checkout {HEAD_BRANCH}
   

3. Verify you are on the correct branch and it is based on origin/{BASE_BRANCH}:

   git log --oneline -1 origin/{BASE_BRANCH}
   git log --oneline -1 HEAD
   

   These should show the same commit (or HEAD should be ahead by only your own commits).

Submodule (mono/skia):

4. Enter the submodule:

   cd externals/skia
   

5. Align to the SHA that origin/{BASE_BRANCH} expects (the submodule tracks {SKIA_BASE_BRANCH} in mono/skia — skiasharp for a main update, or the matching release/<major>.{TARGET}.x for a release-line update — NOT main):

   BASE_SUB_SHA=$(git -C ../.. ls-tree origin/{BASE_BRANCH} -- externals/skia | awk '{print $3}')
   git fetch origin {SKIA_BASE_BRANCH}
   git checkout "$BASE_SUB_SHA"
   

   Verify this SHA is on origin/{SKIA_BASE_BRANCH}:

   git branch -r --contains "$BASE_SUB_SHA" | grep 'origin/{SKIA_BASE_BRANCH}'
   

6. Create the submodule feature branch:

   git checkout -b {HEAD_BRANCH}
   

⚠️ Do NOT skip the SHA alignment step (step 5). If the submodule is at a different SHA than origin/{BASE_BRANCH} expects, the merge will produce phantom diffs — functions that already exist on the base branch will appear as new or removed.

🛑 GATE: Both branches created. Verify:

# In parent repo:
git rev-parse --abbrev-ref HEAD          # → {HEAD_BRANCH}
git merge-base HEAD origin/{BASE_BRANCH}  # → should match origin/{BASE_BRANCH} tip
# In submodule:
git -C externals/skia rev-parse --abbrev-ref HEAD  # → {HEAD_BRANCH}

Phase 5: Upstream Merge (mono/skia)

You should still be inside externals/skia from Phase 4.

1. Merge upstream — use --no-commit for manual conflict resolution:

   git merge --no-commit upstream/{UPSTREAM_REF}   # chrome/m{TARGET}, or main in tip mode
   

2. Resolve conflicts — each conflict must be resolved individually. Never use git merge -s ours or git read-tree --reset — this destroys git blame attribution.

   ⚠️ MANDATORY: classify every fork patch as upstreamed or not before resolving. For each conflicted file, list the fork patches touching it and check whether upstream already carries each one (see gotcha #15 for the exact commands):

   git log --oneline {SKIA_BASE_BRANCH} -- <conflicted-file>           # our fork patches on this file
   git log -S "<distinctive code>" --oneline upstream/{UPSTREAM_REF}   # did upstream adopt it?
   

   • Upstreamed → take upstream's (possibly refined) form; record "<subject>" upstreamed as <sha>.
   • Not upstreamed → re-apply our patch on top of upstream's edits; never drop it; record re-applied.
   • Never blanket git checkout --theirs/--ours on a file you have not classified.

   File Category	Strategy
   BUILD.gn	Combine both — keep upstream structure AND SkiaSharp's platform flags + skiasharp_build target
   DEPS	Combine — keep our dependency pins, accept upstream structure
   RELEASE_NOTES.md, infra/	Take upstream
   C API (include/c/, src/c/)	Keep SkiaSharp — adapt includes/API calls in post-merge commits
   Other upstream source (src/, include/)	Verify-upstream-or-reapply — see gotcha #15

   Audit (mandatory). Snapshot fork patches before merging, then cross-reference every conflict:

   MB=$(git merge-base {SKIA_BASE_BRANCH} upstream/{UPSTREAM_REF})
   git log --oneline "$MB..{SKIA_BASE_BRANCH}" > /tmp/fork-patches-before.txt
   

   For every conflicted file, the fork patch(es) touching it must appear in the mono/skia PR's "Conflicts resolved" table as upstreamed or re-applied. A fork patch on a conflicted file that is neither is a lost patch — STOP and fix it. (Patches whose files did not conflict merge cleanly and need no listing.)

3. Commit the merge:

   git commit  # Creates proper two-parent merge
   

4. Verify our C API files survived the merge:

   ls src/c/*.cpp include/c/*.h  # All files should still exist
   

5. Source file verification — Check for added/deleted upstream files:

   git diff $(git merge-base {SKIA_BASE_BRANCH} upstream/{UPSTREAM_REF})..upstream/{UPSTREAM_REF} --diff-filter=AD --name-only -- src/ include/
   

   Cross-reference against BUILD.gn — new source files may need to be added.

🛑 GATE: Merge complete, conflicts resolved. Verify:

ls src/c/*.cpp include/c/*.h                    # C API files intact
git diff --check                                  # Zero conflict markers
git blame src/c/sk_canvas.cpp | head -20         # Attribution shows original commits, not just merge

✅ Before proceeding to C (Update & Build):

• Parent branch is based on origin/{BASE_BRANCH}
• Submodule is at the SHA referenced by the parent's origin/{BASE_BRANCH} submodule pointer
• Upstream merge committed with proper two-parent history
• C API files intact, zero conflict markers

---

C. Update & Build (Phases 6–7)

Phase 6: Update SkiaSharp Version Files

⚠️ This MUST be done before any native build. The build scripts verify version consistency — if VERSIONS.txt still says the old milestone, the build will fail.

Note: The script automatically resets SK_C_INCREMENT to 0 when the milestone changes. If you had a pending increment that must survive, capture it before running.

📋 This phase is handled by a script. The script updates VERSIONS.txt, cgmanifest.json, azure-templates-variables.yml, and verifies SK_C_INCREMENT — then runs the mandatory verification greps. It exits non-zero if any stale references remain.

In the SkiaSharp parent repo, run:

cd ../..  # back to parent repo (Phase 5 ends inside externals/skia)
pwsh .agents/skills/update-skia/scripts/update-versions.ps1 -Current {CURRENT} -Target {TARGET}

The script handles all of these (so you don't have to do them manually):

• scripts/VERSIONS.txt: milestone, increment→0, soname, assembly, file, ALL ~30 nuget lines
• cgmanifest.json: commitHash, version, chrome_milestone, upstream_merge_commit
• scripts/azure-templates-variables.yml: SKIASHARP_VERSION (must match VERSIONS.txt nuget version)
• Verifies SK_C_INCREMENT is 0 in externals/skia/include/c/sk_types.h
• Runs mandatory grep verification — fails if any stale references remain

🛑 GATE: Script exits with ✅. If it exits with ❌, fix the reported stale references and re-run until it passes.

Note: The SK_C_INCREMENT reset modifies a file in the submodule (externals/skia/). Don't commit it separately — it will be committed with Phase 7's C API fixes.

Phase 7: Fix C API Shim & Build Native

This is where most of the work happens. The C API (src/c/, include/c/) wraps Skia C++ and must be updated when the underlying C++ APIs change.

❌ NEVER use externals-download during a milestone update. It downloads pre-built binaries from the OLD milestone that don't contain your C API changes. Always build from source with externals-{platform}.

1. Restore tools and attempt to build to identify all compilation errors:

   dotnet tool restore
   dotnet cake --target=externals-{platform} --arch={arch}
   

   Replace {platform} with your OS (macos, linux, windows) and {arch} with your architecture (arm64, x64).

2. Fix each error following these patterns:

   Error Type	Fix Pattern
   Missing type	Add/update typedef in sk_types.h
   Renamed function	Update call in *.cpp
   Removed enum value	Remove from sk_enums.cpp + sk_types.h. Note this for Phase 9 — it needs [Obsolete] or documented removal
   Changed signature	Update C wrapper function signature
   New header required	Add #include in the relevant .cpp
   Legacy flag breaks C API	Update C API to use replacement API (see gotcha #6). Do not just comment out the flag without a plan
   New required upstream gn arg	A new upstream dependency our fork doesn't vendor may need a gn toggle (e.g. skia_use_partition_alloc=false). Add it to the affected platforms' native/**/build.cake gn-args lists — NOT a one-off --gnArgs flag (see gotcha #23)

   GN args belong in build.cake, not CLI flags. When the upstream merge introduces a genuinely required new gn argument (typically a dependency our DEPS deliberately doesn't vendor), add it to every affected platform's native/**/build.cake gn-args list — that file is the single source of truth, next to the existing skia_use_* toggles. Don't paper over it with a one-off dotnet cake … --gnArgs flag (non-durable), and don't add gn args (or change compiler/linker flags) merely to silence a host-specific build error — that's a missing-dependency problem, not a config one. Full rationale + the skia_use_partition_alloc example and the milestone-sequencing caveat: gotcha #23.

3. Update sk_types.h for any new enums or type changes. Phase 6 reset SK_C_INCREMENT to 0. Only bump it if you add new C API functions in this milestone. The build enforces that SK_C_INCREMENT matches libSkiaSharp increment in VERSIONS.txt.

4. Build again — iterate until clean compilation.

🛑 GATE: Native library builds successfully on at least one platform.

✅ Before proceeding to D (Regenerate & Verify):

• Version files updated (Phase 6 script passed)
• Native library builds cleanly

---

D. Regenerate & Verify (Phases 8–10)

Phase 8: Regenerate Bindings

Prerequisite: Phase 7's native build must have completed at least once — it runs git-sync-deps, which fetches HarfBuzz and other headers the generator needs.

📋 This phase is handled by a script. The script runs the generator, IMMEDIATELY reverts HarfBuzz bindings (HarfBuzz updates are always separate), reports what changed, and lists any new functions that may need C# wrappers.

pwsh .agents/skills/update-skia/scripts/regenerate-bindings.ps1

The script handles all of these (so you don't forget any):

• Runs pwsh ./utils/generate.ps1
• Reverts binding/HarfBuzzSharp/HarfBuzzApi.generated.cs (proactively, not reactively)
• Reports the binding diff summary
• Lists NEW generated functions that may need C# wrappers in Phase 9

After the script completes, build C# to verify compilation:

dotnet build binding/SkiaSharp/SkiaSharp.csproj

🛑 GATE: Script prints ✅ Phase 8 complete. C# build succeeds with 0 errors.

Phase 9: Fix C# Wrappers

The C# build can pass with 0 errors while new C API functions remain invisible to users. New functions compile fine as unused internal static methods in the generated file, but without C# wrappers they're not part of the public API. This phase applies even when the build succeeds.

1. Review new generated bindings for unwrapped functions:

   git diff origin/{BASE_BRANCH} -- binding/SkiaSharp/SkiaApi.generated.cs | grep "^+.*internal static"
   

   ⚠️ The git diff origin/{BASE_BRANCH} may show additional changes beyond new functions (e.g. struct renames, type changes from Phase 7 shim work). These are expected and correct. Only investigate +internal static lines — ignore other diff noise.

2. Check whether each new function has a C# wrapper:

   # Example: if sk_foo_bar was added, check for a wrapper
   grep -rn "sk_foo_bar" binding/SkiaSharp/*.cs | grep -v generated
   

   New functions from our custom C API additions typically need wrappers. New functions from upstream changes are usually additive and can be deferred.

3. Fix files in binding/SkiaSharp/ based on the breaking change analysis:

   File	When to Update
   Definitions.cs	New enums, types, or constants
   EnumMappings.cs	New enum values that need C#↔C mapping
   GRDefinitions.cs	Graphics context changes (Ganesh)
   SKImage.cs	SkImage factory changes
   SKTypeface.cs	SkTypeface API changes
   SKFont.cs	SkFont API changes
   SKCanvas.cs	Canvas drawing API changes

Key rules:

• Add new overloads, never modify existing signatures (ABI stability)
• Use [Obsolete] for deprecated APIs with migration guidance
• Return null from factory methods on failure (don't throw)

Phase 10: Build & Test

# Rebuild native only if you touched C API files in Phase 9
# (Phase 7 already built — skip if no native changes since then)
dotnet cake --target=externals-{platform} --arch={arch}

# Build C#
dotnet build binding/SkiaSharp/SkiaSharp.csproj

1. Smoke tests (fast gate, ~100ms):

   dotnet test tests/SkiaSharp.Tests.Console/SkiaSharp.Tests.Console.csproj --filter "Category=Smoke"
   

   Smoke tests verify basic native interop: version compatibility, object creation, drawing, image loading, fonts, codecs, effects, and more. If these fail, something fundamental is broken — go back and fix before wasting time on the full suite.

   ⚠️ If the version compatibility smoke test fails with "incompatible native library", you missed a version update — go back to Phase 6 and verify ALL version lines. Do NOT work around this with --no-incremental or by copying native libs manually.

2. Full test suite (required before any PR): capture the output to a log you can inspect afterward. Standalone runs can use any writable path; the automated workflow overrides this to /tmp/gh-aw/agent/test-output.txt so it's uploaded as an artifact (see the workflow's Phase 10 note).

   dotnet test tests/SkiaSharp.Tests.Console/SkiaSharp.Tests.Console.csproj 2>&1 | tee /tmp/skia-test-output.txt
   

   Wait for it to finish (takes 5–7 min). Then read the summary:

   tail -5 /tmp/skia-test-output.txt
   

   The last line will look like: Passed! - Failed: 0, Passed: 5435, Skipped: 171, Total: 5606

   This runs all test projects (core, Vulkan, Direct3D). Backend-specific tests self-skip when hardware isn't available. CI handles WASM/Android/iOS separately.

   ⚠️ These MUST be two separate commands. Do NOT combine them into a single pipeline like | tee ... | tail — the piped tail runs immediately and will show nothing useful while tests are still running. Capture with tee first, wait for completion, then tail the output file. After the run, inspect failures with:

   grep '^  Failed' /tmp/skia-test-output.txt
   

Smoke tests are just that — smoke. They verify the basics. The full suite MUST pass before the update can be considered complete. Do not create PRs with only smoke tests passing.

🛑 GATE: ALL tests pass (full suite, not just smoke). Do NOT skip failing tests. Do NOT proceed with failures.

✅ Before proceeding to E (Ship):

• Bindings regenerated (Phase 8 script passed)
• C# builds with 0 errors
• ALL tests pass (full suite)

---

E. Ship (Phase 11)

Phase 11: Create PRs

Same-milestone bug-fix syncs: When CURRENT == TARGET, only cgmanifest.json's commitHash/upstream_merge_commit change. Use PR titles like [skia-sync] Merge upstream chrome/m{TARGET} bug fixes instead of milestone-bump titles. A release-line update is always such a sync.

Branch targets by mode: Use the {BASE_BRANCH} / {SKIA_BASE_BRANCH} / {HEAD_BRANCH} values resolved in Phase 1 step 5. For a main update they are main / skiasharp / skia-sync/m{TARGET}; for main/tip mode main / skiasharp / skia-sync/main; for a release-line update release/<major>.{TARGET}.x (both repos) / skia-sync/release-<major>.{TARGET}.x.

PR 1: mono/skia (submodule)

Field	Value
Branch	{HEAD_BRANCH}
Target	{SKIA_BASE_BRANCH}
Title	[skia-sync] Merge upstream chrome/m{TARGET}

PR 2: mono/SkiaSharp (parent)

Field	Value
Branch	{HEAD_BRANCH}
Target	{BASE_BRANCH}
Title	[skia-sync] Update skia to milestone {TARGET} (or Merge upstream chrome/m{TARGET} bug fixes when CURRENT == TARGET)

Submodule must point to the mono/skia PR branch.

Cross-link both PRs (same as native-dependency-update).

After creating BOTH PRs, update the earlier PR's description to include a link to the later one. Both PRs must reference each other.

Phase 11 Completion Checklist

Before proceeding to merge, verify ALL of these:

• Branch names follow {HEAD_BRANCH} convention in BOTH repos
• mono/skia PR targets {SKIA_BASE_BRANCH} branch
• mono/SkiaSharp PR targets {BASE_BRANCH} branch
• SkiaSharp's externals/skia submodule points to the mono/skia PR branch (git submodule status)
• cgmanifest.json updated with new commit hash, version, and chrome_milestone
• scripts/VERSIONS.txt updated (ALL version lines, not just milestone)
• SkiaApi.generated.cs regenerated and committed
• Both PRs cross-reference each other
• Native build passes on at least one platform
• C# build passes with 0 errors
• All tests pass (with proper platform filtering)

Merge Sequence (CRITICAL)

1. Merge mono/skia PR first → creates new squashed SHA on {SKIA_BASE_BRANCH}
2. Fetch new SHA in SkiaSharp's submodule
3. Update submodule pointer, push to SkiaSharp PR branch
4. Only then merge SkiaSharp PR

Merge Checklist

Before proceeding past each step, verify:

• mono/skia PR merged
• Fetched {SKIA_BASE_BRANCH} branch to get new squashed SHA
• Updated SkiaSharp submodule to new SHA (cd externals/skia && git fetch origin && git checkout {new-sha})
• Pushed submodule update to SkiaSharp PR branch
• CI passes on updated SkiaSharp PR
• SkiaSharp PR merged
• Submodule points to a commit on {SKIA_BASE_BRANCH} branch (not an orphaned branch commit)

❌ NEVER merge both PRs without updating the submodule in between. ❌ NEVER assume the submodule reference is correct after squash-merging mono/skia.

---

Reference Material

These files contain lookup information — consult them when you hit a problem or need context, not necessarily upfront:

• references/known-gotchas.md — Hard-won lessons from past updates (DEF_STRUCT_MAP, emsdk, BUILD.gn flags, HarfBuzz, DEPS forks, etc.) and a troubleshooting table
• references/typical-changes.md — Files typically changed in each repository during an update
• references/breaking-changes-checklist.md — How to analyze breaking changes between milestones