Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.
Claude CodeNot declared
CursorNot declared
CodexNot declared
GitHub CopilotNot declared
WindsurfNot declared
Gemini CLINot declared
ClineNot declared
OpenCodeNot declared
Repository health
Stars181
LicenseLICENSE
Default branchmaster
Open issues0
Status
Archived
Package contents
Files included with this skill beyond the listing page.
skill mdSKILL.md4,465 B
docsSUMMARY.md454 B
History
First seen on skills.sh
First recorded snapshot · 4 installs
SKILL.md
Android Memory Remediate
Turn a supported Android memory diagnosis into the smallest architecture-correct change and prove its effect in the same scenario.
Entry Gate
Require a diagnosis with artifact IDs, accounting domain, owner, mechanism, and residual alternatives.
If the diagnosis revises a prior analysis, use only the current confirmed or newly supported owner/mechanism as the entry gate. A retracted, unresolved, or merely inherited prior claim cannot authorize a code change.
Treat QA logs and screenshots as admissible evidence only when the diagnosis binds the exact line/hash or visible region to the matching target/build/phase. An OOM line, keyword match, notification, or chart screenshot alone does not open the code-change gate.
If evidence support is insufficient or the owner is unresolved, invoke $android-memory-evidence and $android-memory-diagnose first.
Permit a collection/instrumentation change without a root cause when its explicit purpose is to obtain the missing discriminator.
Never implement a generic memory tip merely because a threshold fired.
Confirm the target source root, module, revision, and build variant. When a workspace contains both an analyzer and a demo/test fixture, do not infer the edit target from package name alone.
Workflow
Read the target project's instructions, source, tests, architecture, current branch, and worktree before editing.
Confirm graphics changes preserve rendering correctness and release ordering.
Confirm device/API/vendor branches are capability-gated rather than model-name hardcoded.
Confirm observability distinguishes unsupported, permission denied, empty, and true absence.
Handoff
Return:
evidence-backed root cause or instrumentation purpose;
changed files and architecture rationale;
tests and project validation results;
before/after memory evidence with matching accounting domain;
product/performance regressions checked;
residual uncertainty and missing device validation;
rollback condition.
If the Entry Gate fails, return a blocked handoff instead: facts currently supported, unresolved owner/mechanism or target-project fields, the smallest discriminator and its prerequisites, code and files deliberately left unchanged, and the condition that would permit implementation. Do not make the user infer that stopping was intentional.