google · LalitMaganti · May 16, 2026
@@ -44,22 +44,22 @@ without any flattening step.
 ```
 ai/skills/
 ├── README.md
-├── perfetto-infra-querying-traces/
+├── perfetto_infra_querying_traces/
 │   └── SKILL.md
-├── perfetto-infra-getting-trace-processor/
+├── perfetto_infra_getting_trace_processor/
 │   └── SKILL.md
-└── perfetto-workflow-android-heap-dump/
+└── perfetto_workflow_android_heap_dump/
     └── SKILL.md
 ```
 
-The slug doubles as the taxonomy: `perfetto-` is the vendor prefix
+The slug doubles as the taxonomy: `perfetto_` is the vendor prefix
 (so Perfetto skills don't collide with anything else the user has
 installed), then a category, then any sub-categories, then the
 skill name. Two categories today:
 
-- **`perfetto-infra-*`** — domain-agnostic mechanics of working with
+- **`perfetto_infra_*`** — domain-agnostic mechanics of working with
   Perfetto: how to query a trace, how to install the binary, etc.
-- **`perfetto-workflow-<domain>-*`** — domain-specific
+- **`perfetto_workflow_<domain>_*`** — domain-specific
   investigation workflows: how to investigate a heap dump on
   Android, jank on Chrome, etc.
 
@@ -69,11 +69,11 @@ need to query the trace; here's the bit specific to this problem."
 ## Authoring a skill
 
 Create a directory under `ai/skills/` whose name follows the
-`perfetto-<category>-<name>` pattern, with a `SKILL.md` inside:
+`perfetto_<category>_<name>` pattern, with a `SKILL.md` inside:
 
 ```markdown
 ---
-name: perfetto-infra-querying-traces
+name: perfetto_infra_querying_traces
 description: Use when the user wants to load a Perfetto trace, run
   a SQL query against it, or discover which tables and columns are
   available. Covers trace_processor invocation and the PerfettoSQL

@@ -21,7 +21,7 @@ perfetto_validate_skill_test(
 )
 
 perfetto_agent_skill(
-    name = "perfetto-infra-getting-trace-processor",
+    name = "perfetto_infra_getting_trace_processor",
     srcs = ["SKILL.md"] + glob(["references/**", "assets/**"]),
     scripts = glob(["scripts/**"]),
 )
@@ -2,7 +2,7 @@
 # proto-message: EvalSuite
 # txtpbfmt: wrap_strings_after_newlines
 
-suite_name: "perfetto-infra-getting-trace-processor"
+suite_name: "perfetto_infra_getting_trace_processor"
 timeout_seconds: 300
 
 cases {

@@ -1,5 +1,5 @@
 ---
-name: perfetto-infra-getting-trace-processor
+name: perfetto_infra_getting_trace_processor
 description: Use when the user does not yet have `trace_processor`
   installed, or has not set up the Python `perfetto` client used to
   drive the long-running RPC mode. Covers fetching the prebuilt binary

@@ -1,11 +1,11 @@
-# TEST for perfetto-infra-getting-trace-processor
+# TEST for perfetto_infra_getting_trace_processor
 
 ## Tier 1 — Build & Smoke Tests
 
 ### Test 1: File Existence
 Verify that the skill files exist.
 ```bash
-ls ai/skills/perfetto-infra-getting-trace-processor/SKILL.md
+ls ai/skills/perfetto_infra_getting_trace_processor/SKILL.md
 ```
 **Verify:** File exists.
 

@@ -21,7 +21,7 @@ perfetto_validate_skill_test(
 )
 
 perfetto_agent_skill(
-    name = "perfetto-infra-querying-traces",
+    name = "perfetto_infra_querying_traces",
     srcs = ["SKILL.md"] + glob(["references/**", "assets/**"]),
     scripts = glob(["scripts/**"]),
 )
@@ -2,7 +2,7 @@
 # proto-message: EvalSuite
 # txtpbfmt: wrap_strings_after_newlines
 
-suite_name: "perfetto-infra-querying-traces"
+suite_name: "perfetto_infra_querying_traces"
 timeout_seconds: 300
 
 cases {

@@ -1,5 +1,5 @@
 ---
-name: perfetto-infra-querying-traces
+name: perfetto_infra_querying_traces
 description: Use when the user wants to load a Perfetto trace, run a
   PerfettoSQL query against it, or discover which tables, views, columns,
   or stdlib modules are available. Covers trace_processor invocation, the
@@ -20,7 +20,7 @@ runs on top of, including the Perfetto UI. Reference docs:
 > `trace_processor` is already on `PATH` (and, for the long-running RPC
 > mode below, that the Python client is installed). If neither is set
 > up, follow whatever skill the user's environment provides for
-> acquisition — typically `perfetto-infra-getting-trace-processor` for the
+> acquisition — typically `perfetto_infra_getting_trace_processor` for the
 > open-source path, or a team-specific variant inside Google or other
 > restricted environments. The rest of this skill is intentionally
 > orthogonal to *how* you got `trace_processor`.
@@ -166,7 +166,7 @@ A few commonly used modules to know:
 - `android.startup.startups` — one row per app startup.
 - `stacks.cpu_profiling` — flat samples and call-graph helpers.
 - `android.memory.heap_graph.dominator_tree` — retained-size analysis for
-  Java heap dumps (see the `perfetto-workflow-android-heap-dump`
+  Java heap dumps (see the `perfetto_workflow_android_heap_dump`
   skill for usage).
 
 The module name maps directly to the file path under the stdlib root:

@@ -1,11 +1,11 @@
-# TEST for perfetto-infra-querying-traces
+# TEST for perfetto_infra_querying_traces
 
 ## Tier 1 — Build & Smoke Tests
 
 ### Test 1: File Existence
 Verify that the skill files exist.
 ```bash
-ls ai/skills/perfetto-infra-querying-traces/SKILL.md
+ls ai/skills/perfetto_infra_querying_traces/SKILL.md
 ```
 **Verify:** File exists.
 

@@ -21,7 +21,7 @@ perfetto_validate_skill_test(
 )
 
 perfetto_agent_skill(
-    name = "perfetto-workflow-android-heap-dump",
+    name = "perfetto_workflow_android_heap_dump",
     srcs = ["SKILL.md"] + glob(["references/**", "assets/**"]),
     scripts = glob(["scripts/**"]),
 )
@@ -2,7 +2,7 @@
 # proto-message: EvalSuite
 # txtpbfmt: wrap_strings_after_newlines
 
-suite_name: "perfetto-workflow-android-heap-dump"
+suite_name: "perfetto_workflow_android_heap_dump"
 timeout_seconds: 300
 
 cases {

@@ -1,5 +1,5 @@
 ---
-name: perfetto-workflow-android-heap-dump
+name: perfetto_workflow_android_heap_dump
 description: Use when the user has an Android trace containing a Java heap
   graph (heap dump) and wants to investigate memory usage, find leaks, or
   understand what is retaining memory. Walks through a guided workflow
@@ -15,7 +15,7 @@ with this heap dump?", "why is this process using so much memory?", or
 data source and contains at least one Java heap graph.
 
 If the user has not yet loaded a trace into `trace_processor`, follow
-the `perfetto-infra-querying-traces` skill first, then come back here.
+the `perfetto_infra_querying_traces` skill first, then come back here.
 Recording-side reference for heap dumps:
 <https://perfetto.dev/docs/data-sources/java-heap-profiler>.
 

@@ -1,11 +1,11 @@
-# TEST for perfetto-workflow-android-heap-dump
+# TEST for perfetto_workflow_android_heap_dump
 
 ## Tier 1 — Build & Smoke Tests
 
 ### Test 1: File Existence
 Verify that the skill files exist.
 ```bash
-ls ai/skills/perfetto-workflow-android-heap-dump/SKILL.md
+ls ai/skills/perfetto_workflow_android_heap_dump/SKILL.md
 ```
 **Verify:** File exists.