[vmelikyan]

Reader and task

• Audience: application developers who configure native builds in lifecycle.yaml.
• Task/outcome: choose how BuildKit uses the build cache for a service (docker.builder.cache.mode: max, min or off), and know when min or off makes builds faster.

Source of truth

• Exact owning repository revisions, PR head SHAs, or immutable artifacts: lifecycle bc9dc7dfb5511161e2bf744580815551ebf28561 (feat: per-service build cache mode for native builds lifecycle#266, released in 0.2.3-alpha.45).
• Frontmatter verification baseline added or reused: added 2026-10-21-build-cache-mode-released. Other claims on the edited pages keep 2026-07-24-comprehensive-audit.
• Product or runtime verification: checked against the source of the merged change: the schema enum, the BuildKit export and import flags for each mode, the Kaniko cache flags (removed only for off), and no effect for the ci engine.
• Contract checks (schema / OpenAPI / CLI; identify any SKIP): check:contracts --require-all passed against the change's PR head (same code as the merge): schema 56 examples including the 2 new ones, OpenAPI 41, CLI 96. The contract step inside bun run verify is a SKIP.
• Unavailable or unresolved sources: the install-wide default (buildDefaults.cache.mode) is not documented, the same as the other buildDefaults keys. The pages say only that Lifecycle uses the installation default, which is max unless an administrator changed it.

Documentation changes

• Changed pages: /docs/schema/github (field table row, new docker.builder.cache section with an example, cache.mode in the "All fields" example) and /docs/schema/helm (field table row and a link to the GitHub section).
• ASD-STE100 human review: every changed sentence reviewed against the 53 rule rows; style baseline updated after the review.
• Navigation or redirects: none.
• Raw Markdown projection: public/llms.txt regenerated with bun run build:llms.
• External state: none.
• Screenshots, catalog status, and manual pixel/state review: no visual changes. The new section, table rows and the helm-to-github anchor render and resolve in the static build.

Verification

• Formatted changed files
• Kept source-review evidence and non-actionable implementation details out of end-user prose
• Confirmed that each changed sentence helps the reader understand, complete, verify, or recover the task
• Reviewed all 53 rule rows in .agents/skills/update-lifecycle-docs/references/ste100-review.md
• bun run check:styles
• Updated the style baseline only after the complete human review
• bun run check:raw
• bun run verify
• bun run check:contracts --require-all when schema, API, or CLI claims changed
• Checked changed routes with an approved desktop browser inspection tool
• Checked every theme enabled by the current theme config (dark only: the static build rendered dark even with light mode forced)
• Ran bun run check:screenshots when visuals changed (no visual changes)
• Confirmed screenshots contain no secrets or personal/private data
• Confirmed tracked guidance has no contributor-specific paths, hosts, credentials, or fixture identities