Neta Skill Services 100x 综合评测报告

1. Executive Decision

用户提出的硬性要求已经满足：每个 line::interface 都至少测了 100 次。这个要求已经不是推断，而是直接由聚合数据验证。

顶层结论没有因为样本扩大而变好，反而更清楚了。整体仍然不适合当作“可放心交付”的完整技能服务集来推广。

CN 的主要问题是写和创作几乎全面失效，很多调用是“看起来执行了，但没有产生真实结果”。
COM 的主要问题是 public-read 链路很慢、超时多、detail/read 类接口反复失真。
Naturalness 仍然不够，新的 agent 拿到后并不能自然地首步做对、也不能稳定避免隐性知识依赖。

2. Coverage

Coverage Metric	Value
Interface dimensions	50
Minimum attempts per dimension	100
Maximum attempts per dimension	294
Dimensions below 100	0
Dimensions exactly 100	43

Evidence Metric	Value
Interface run shards	20
Naturalness runs	2
Cases	237
Attempts	5,217
Attempts per case avg	22.013

接口 100x 这条线已经完成，但 naturalness 仍然来自单独的 16 个 cold-start case。它不是 100x 套件的一部分，所以报告里单独标出，避免混口径。

full-20260321-formal, full-20260321-formal-r2, soak-smoke-cn-cap1, soak-full-cn-100x-r3, soak-full-com-100x-r3, soak-full-com-tail-r1, quick-run, read-run, read-run-2, read-run-3, feed-run, search-run, request-run, adventure-run, image-bg-run, image-video-run, image-video-run-2, image-video-run-3, image-video-run-4, image-video-run-5

The invalid standalone shard `full-com-video-r1` is intentionally excluded from the 100x aggregate because it lacked the setup context required for a meaningful `make_video` attempt.

3. Scoring Contract

这套评测不是看 shell 是否退出成功，而是看目标能力是否真的完成。只要返回不完整、需要的 artifact 没有落地、或者 CLI 把失败伪装成成功，都会被判成非成功。

Effective success: 真正拿到可用结果，后续链路可继续。
False success: 表面完成，但真实产物或真实效果不存在。
Timeout: 达到超时门槛，哪怕之前有部分输出也不算成功。
Naturalness: 新 agent 是否能自然选线、自然首步、自然理解接口语义。

口径说明： 报告里的 false_success_count 是 reviewer 级别的质量标记，不等于终态 result_class=false_success 的数量。一个调用可能最终被记成 auth_error 或 dependency_missing，但如果中间输出曾误导性地“看起来成功”，仍会被记入 false-success。

4. Aggregate Results

4.1 Overall

Metric	Value
Cases	237
Attempts	5,217
Effective success count	2627
Effective success rate	0.504
Effective success case rate	0.646
False success count	936
Timeout count	106
Retry count total	2
Latency p50 ms	1,384
Latency p95 ms	41,264

4.2 By Line

Line	Attempts	Success Rate	False Success	Timeout	Retry	p95 ms
`cn`	2500	0.349	604	1	0	1,576
`com`	2717	0.646	332	105	2	58,155

4.3 High-Level Read

cn 的速度看起来好，是因为大量失败发生在非常早的阶段。
com 顶层成功更高，但延迟尾部明显失控，尤其 public-read。
5217 次尝试之后，问题已经不是偶发波动，而是稳定的系统性缺陷。

4.4 By Category

Scope	Class	Cases	Attempts	Success Rate	Case Success Rate	False Success	Timeout	Retry	p95 ms
`cn::public_read`	conditional	80	1400	0.624	0.825	0	1	0	1,657
`cn::authenticated_write`	fail	31	700	0.000	0.000	404	0	0	1,358
`cn::heavy_creative`	fail	16	400	0.000	0.000	200	0	0	1,445
`com::public_read`	fail	69	1407	0.547	0.739	104	103	2	40,017
`com::authenticated_write`	conditional	25	701	0.859	0.800	3	0	0	3,668
`com::heavy_creative`	fail	16	609	0.627	1.000	225	2	0	79,543

最关键的结构性判断： com authenticated_write 是唯一接近可交付的切片，但仍未达到严格 gate； cn authenticated_write、cn heavy_creative、com public_read、com heavy_creative 都不能作为放心上线面来对外承诺。

5. Interface Hotspots

5.1 Worst Success Rate

Scope	Category	Attempts	False Success	Timeout	p95 ms
`cn::create_adventure_campaign`	`authenticated_write`	100	100	0	1,305
`cn::list_my_adventure_campaigns`	`authenticated_write`	100	100	0	1,333
`cn::list_my_characters`	`authenticated_write`	100	100	0	1,439
`cn::list_my_elementum`	`authenticated_write`	100	100	0	1,399
`cn::make_image`	`heavy_creative`	100	100	0	1,537
`cn::make_song`	`heavy_creative`	100	100	0	1,424
`com::request_interactive_feed`	`public_read`	100	50	0	3,668
`cn::like_collection`	`authenticated_write`	100	4	0	0
`com::read_collection`	`public_read`	103	0	103	40,025
`cn::make_video`	`heavy_creative`	100	0	0	0

5.2 Highest False Success

Scope	Category	Attempts	Success Rate	False Success	p95 ms
`com::make_video`	`heavy_creative`	294	0.289	209	76,941
`cn::create_adventure_campaign`	`authenticated_write`	100	0.000	100	1,305
`cn::list_my_adventure_campaigns`	`authenticated_write`	100	0.000	100	1,333
`cn::list_my_characters`	`authenticated_write`	100	0.000	100	1,439
`cn::list_my_elementum`	`authenticated_write`	100	0.000	100	1,399
`cn::make_image`	`heavy_creative`	100	0.000	100	1,537
`cn::make_song`	`heavy_creative`	100	0.000	100	1,424
`com::request_interactive_feed`	`public_read`	100	0.000	50	3,668
`com::request_character_or_elementum`	`public_read`	100	0.500	50	3,732
`com::remove_background`	`heavy_creative`	100	0.880	12	19,971

5.3 Highest Timeout

Scope	Category	Attempts	Success Rate	False Success	Timeout	p95 ms
`com::read_collection`	`public_read`	103	0.000	0	103	40,025
`com::make_song`	`heavy_creative`	109	0.972	1	2	110,379
`cn::search_character_or_elementum`	`public_read`	100	0.990	0	1	1,588
`cn::create_adventure_campaign`	`authenticated_write`	100	0.000	100	0	1,305
`cn::like_collection`	`authenticated_write`	100	0.000	4	0	0
`cn::list_my_adventure_campaigns`	`authenticated_write`	100	0.000	100	0	1,333
`cn::list_my_characters`	`authenticated_write`	100	0.000	100	0	1,439
`cn::list_my_elementum`	`authenticated_write`	100	0.000	100	0	1,399
`cn::make_image`	`heavy_creative`	100	0.000	100	0	1,537
`cn::make_song`	`heavy_creative`	100	0.000	100	0	1,424

Stable Examples

CN stable subset: list_spaces, request_interactive_feed, suggest_categories, suggest_content, suggest_keywords, suggest_tags, validate_tax_path
COM stable subset: create_adventure_campaign, list_my_adventure_campaigns, list_my_characters, list_my_elementum, request_adventure_campaign, list_spaces, search_character_or_elementum, suggest_categories, suggest_content, suggest_keywords, suggest_tags, validate_tax_path

Priority Repair Surfaces

cn::create_adventure_campaign, cn::list_my_*, cn::make_image, cn::make_song
com::read_collection, com::request_interactive_feed, com::request_character_or_elementum
com::make_video is the single largest false-success sink

5.4 Interface / Action / Data Map

这张表把接口、动作、输入依赖、成功判据、导出数据和下游依赖放到一张图里。读这张表时，不需要再去反推 case 文件。

Interface	User Action	Inputs	Success Evidence	Exported Data	Downstream Use	CN Health	COM Health	Priority
`create_adventure_campaign`	Create a new draft adventure campaign	`adventure_name` from line profile seed `mission_plot` from line profile seed `mission_rules` from line profile seed `mission_task` from line profile seed	`status` `uuid`	`campaign_uuid` <= uuid	`request_adventure_campaign` `update_adventure_campaign`	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	healthy healthy \| attempts 101 \| success 1.000 \| false 0 \| timeout 0	critical
`get_hashtag_characters`	Browse characters under a topic hashtag	`topic_hashtag` from `list_space_topics`	`total`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical
`get_hashtag_collections`	Browse collections under a space or topic hashtag	`main_hashtag` from `list_spaces` `topic_hashtag` from `list_space_topics`	`total`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.080 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.040 \| false 4 \| timeout 0	critical
`get_hashtag_info`	Read a space or hashtag detail card	`main_hashtag` from `list_spaces`	`hashtag.name`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical
`like_collection`	Like or unlike a collection	`collection_uuid` from `suggest_content`	`success`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 4 \| timeout 0	critical critical \| attempts 100 \| success 0.020 \| false 2 \| timeout 0	critical
`list_my_adventure_campaigns`	List current user's adventure campaigns	-	`total`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	critical
`list_my_characters`	List current user's created characters	-	`total`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	critical
`list_my_elementum`	List current user's created elementa	-	`total`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	critical
`list_space_topics`	Expand a space into sub-topics	`space_uuid` from `list_spaces`	`topics.primary_topic.hashtag_name`	`topic_hashtag` <= topics.primary_topic.hashtag_name, topics.topics[0].hashtag_name	`get_hashtag_characters` `get_hashtag_collections`	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.040 \| false 0 \| timeout 0	critical
`list_spaces`	Discover available spaces and world entries	-	`spaces[0].name` `spaces[0].space_uuid`	`main_hashtag` <= spaces[0].main_hashtag_name, spaces[0].name `space_uuid` <= spaces[0].space_uuid	`get_hashtag_collections` `get_hashtag_info` `list_space_topics`	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy
`make_image`	Generate an image artifact	`image_prompt` from line profile seed	`artifacts[0].uuid` `task_uuid`	`image_artifact_uuid` <= artifacts[0].uuid `image_url` <= artifacts[0].url	`make_video` `remove_background`	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	high high \| attempts 106 \| success 0.972 \| false 3 \| timeout 0	critical
`make_song`	Generate song and lyric artifacts	`song_lyrics` from line profile seed `song_prompt` from line profile seed	`artifacts[0].audio_detail.lyric_url` `task_uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 100 \| timeout 0	high high \| attempts 109 \| success 0.972 \| false 1 \| timeout 2	critical
`make_video`	Generate video from an image source	`image_url` from `make_image` `video_prompt` from line profile seed	`artifacts[0].detail_url` `task_uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 0 \| timeout 0	critical critical \| attempts 294 \| success 0.289 \| false 209 \| timeout 0	critical
`read_collection`	Open a concrete collection detail page	`collection_uuid` from `suggest_content`	`collection.uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.020 \| false 0 \| timeout 0	critical critical \| attempts 103 \| success 0.000 \| false 0 \| timeout 103	critical
`remove_background`	Remove image background from a generated image	`image_artifact_uuid` from `make_image`	`artifacts[0].uuid` `task_uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 0 \| timeout 0	high high \| attempts 100 \| success 0.880 \| false 12 \| timeout 0	critical
`request_adventure_campaign`	Read back one adventure campaign	`campaign_uuid` from `create_adventure_campaign`	`name` `uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	critical
`request_character_or_elementum`	Read character or elementum detail	`character_uuid_top` from `search_character_or_elementum` `keyword_element` from line profile seed	`detail.name` `detail.uuid`	`No exported variable; only response payload is validated`	-	high high \| attempts 100 \| success 0.530 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.500 \| false 50 \| timeout 0	critical
`request_interactive_feed`	Scroll interactive feed pages with session trace	`biz_trace_id` from `request_interactive_feed`	`module_list[0].json_data.uuid`	`biz_trace_id` <= page_data.biz_trace_id	`request_interactive_feed`	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	critical critical \| attempts 100 \| success 0.000 \| false 50 \| timeout 0	critical
`search_character_or_elementum`	Search characters or elementum by keyword	`keyword_char` from line profile seed `keyword_element` from line profile seed	`list[0].uuid` `total`	`character_uuid_top` <= list[0].uuid	`request_character_or_elementum`	high high \| attempts 100 \| success 0.990 \| false 0 \| timeout 1	healthy healthy \| attempts 101 \| success 1.000 \| false 0 \| timeout 0	high
`suggest_categories`	Navigate the 3-level taxonomy tree	`primary_category` from `suggest_categories` `primary_category>$secondary_category` from upstream runtime context	`suggestions[0].name`	`primary_category` <= suggestions[0].name `secondary_category` <= suggestions[0].name `tertiary_category` <= suggestions[0].name	`suggest_categories`	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy
`suggest_content`	Fetch recommend/search/exact content feed	`keyword_tag` from line profile seed `primary_category>$secondary_category>$tertiary_category` from upstream runtime context	`module_list[0].json_data.uuid` `page_data.has_next_page`	`collection_uuid` <= module_list[0].json_data.uuid	`like_collection` `read_collection`	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy healthy \| attempts 103 \| success 1.000 \| false 0 \| timeout 0	healthy
`suggest_keywords`	Get keyword suggestions from a prefix	`keyword_prefix` from line profile seed	`suggestions[0].text`	`No exported variable; only response payload is validated`	-	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy
`suggest_tags`	Get related tags from a keyword	`keyword_tag` from line profile seed	`suggestions[0].name`	`No exported variable; only response payload is validated`	-	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy
`update_adventure_campaign`	Update one field on an existing campaign	`campaign_uuid` from `create_adventure_campaign`	`subtitle` `uuid`	`No exported variable; only response payload is validated`	-	critical critical \| attempts 100 \| success 0.000 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 0.990 \| false 1 \| timeout 0	critical
`validate_tax_path`	Validate a candidate taxonomy path before use	`primary_category>$secondary_category>$tertiary_category` from upstream runtime context	`valid`	`No exported variable; only response payload is validated`	-	healthy healthy \| attempts 100 \| success 0.990 \| false 0 \| timeout 0	healthy healthy \| attempts 100 \| success 1.000 \| false 0 \| timeout 0	healthy

5.5 Priority Problem Interfaces

这是最适合直接拿来排修复优先级的列表。每一项都明确指出失败主因，以及为什么它会影响后续动作链。

Scope	Level	User Action	False Success	Timeout	Dominant Failure	Why It Matters	Suggested Fix
`cn::create_adventure_campaign`	critical	Create a new draft adventure campaign	100	0	`auth_error`	blocks `request_adventure_campaign` blocks `update_adventure_campaign`	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`cn::list_my_adventure_campaigns`	critical	List current user's adventure campaigns	100	0	`auth_error`	direct user-facing endpoint	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`cn::list_my_characters`	critical	List current user's created characters	100	0	`auth_error`	direct user-facing endpoint	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`cn::list_my_elementum`	critical	List current user's created elementa	100	0	`auth_error`	direct user-facing endpoint	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`cn::make_image`	critical	Generate an image artifact	100	0	`auth_error`	blocks `make_video` blocks `remove_background`	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`cn::make_song`	critical	Generate song and lyric artifacts	100	0	`auth_error`	direct user-facing endpoint	Verify token scope, line routing, and auth preconditions before treating the call as usable.
`com::request_interactive_feed`	critical	Scroll interactive feed pages with session trace	50	0	`dependency_missing`	blocks `request_interactive_feed`	Validate upstream exports and fail fast when prerequisite IDs or artifacts are absent.
`cn::like_collection`	critical	Like or unlike a collection	4	0	`dependency_missing`	direct user-facing endpoint	Validate upstream exports and fail fast when prerequisite IDs or artifacts are absent.
`com::read_collection`	critical	Open a concrete collection detail page	0	103	`timeout`	direct user-facing endpoint	Add timeout recovery and pagination/session handling, especially for read/detail flows.
`cn::make_video`	critical	Generate video from an image source	0	0	`dependency_missing`	direct user-facing endpoint	Validate upstream exports and fail fast when prerequisite IDs or artifacts are absent.
`cn::remove_background`	critical	Remove image background from a generated image	0	0	`dependency_missing`	direct user-facing endpoint	Validate upstream exports and fail fast when prerequisite IDs or artifacts are absent.
`cn::request_adventure_campaign`	critical	Read back one adventure campaign	0	0	`dependency_missing`	direct user-facing endpoint	Validate upstream exports and fail fast when prerequisite IDs or artifacts are absent.

6. Naturalness

Naturalness 不是接口是否存在，而是新 agent 是否会“天然理解并正确使用”。这正是用户特别强调的交付要求之一。

Metric	Value
Runs	2
Cases	16
Cold-start success count	9
Cold-start success rate	0.562
Manual hint rate	0.000
Wrong first command rate	0.750
Hidden knowledge dependency rate	0.312

Line	Cases	Cold-start Success	Manual Hint	Wrong First	Hidden Knowledge
`cn`	8	0.500	0.000	0.625	0.125
`com`	8	0.625	0.000	0.875	0.500

最核心的问题不是“agent 需要人提醒”，因为 manual-hint rate 是 0。
真正的问题是 agent 在冷启动时经常第一步就走错，且需要知道一些文档里没自然暴露出来的隐藏知识。
com 的 naturalness 更差，wrong-first-command 达到 0.875。

7. Gate Result

最终 gate 结果为 FAIL，通过了 17/29 项。

Check	Actual	Comparator	Threshold	Pass
`overall.false_success_rate`	0.1794134560092007	<=	0.0	fail
`cn.authenticated_write.timeout_rate`	0.0	<=	0.02	pass
`cn.authenticated_write.p95_latency_ms`	1358.0	<=	10000.0	pass
`cn.authenticated_write.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`cn.authenticated_write.false_success_rate`	0.5771428571428572	<=	0.0	fail
`cn.heavy_creative.timeout_rate`	0.0	<=	0.05	pass
`cn.heavy_creative.p95_latency_ms`	1445.0	<=	90000.0	pass
`cn.heavy_creative.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`cn.heavy_creative.false_success_rate`	0.5	<=	0.0	fail
`cn.public_read.timeout_rate`	0.0007142857142857143	<=	0.0	fail
`cn.public_read.p95_latency_ms`	1657.0	<=	5000.0	pass
`cn.public_read.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`cn.public_read.false_success_rate`	0.0	<=	0.0	pass
`com.authenticated_write.timeout_rate`	0.0	<=	0.02	pass
`com.authenticated_write.p95_latency_ms`	3668.0	<=	10000.0	pass
`com.authenticated_write.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`com.authenticated_write.false_success_rate`	0.0042796005706134095	<=	0.0	fail
`com.heavy_creative.timeout_rate`	0.003284072249589491	<=	0.05	pass
`com.heavy_creative.p95_latency_ms`	79543.0	<=	90000.0	pass
`com.heavy_creative.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`com.heavy_creative.false_success_rate`	0.3694581280788177	<=	0.0	fail
`com.public_read.timeout_rate`	0.07320540156361052	<=	0.0	fail
`com.public_read.p95_latency_ms`	40017.0	<=	5000.0	fail
`com.public_read.retry_budget_exceeded_rate`	0.0	<=	0.05	pass
`com.public_read.false_success_rate`	0.07391613361762615	<=	0.0	fail
`naturalness.cold_start_success_rate`	0.5625	>=	0.8	fail
`naturalness.manual_hint_rate`	0.0	<=	0.2	pass
`naturalness.wrong_first_command_rate`	0.75	<=	0.15	fail
`naturalness.hidden_knowledge_dependency_rate`	0.3125	<=	0.1	fail

Failing Checks

overall.false_success_rate actual 0.1794134560092007 violates <= 0.0
cn.authenticated_write.false_success_rate actual 0.5771428571428572 violates <= 0.0
cn.heavy_creative.false_success_rate actual 0.5 violates <= 0.0
cn.public_read.timeout_rate actual 0.0007142857142857143 violates <= 0.0
com.authenticated_write.false_success_rate actual 0.0042796005706134095 violates <= 0.0
com.heavy_creative.false_success_rate actual 0.3694581280788177 violates <= 0.0
com.public_read.timeout_rate actual 0.07320540156361052 violates <= 0.0
com.public_read.p95_latency_ms actual 40017.0 violates <= 5000.0
com.public_read.false_success_rate actual 0.07391613361762615 violates <= 0.0
naturalness.cold_start_success_rate actual 0.5625 violates >= 0.8
naturalness.wrong_first_command_rate actual 0.75 violates <= 0.15
naturalness.hidden_knowledge_dependency_rate actual 0.3125 violates <= 0.1

8. Full 100x Interface Table

Line	Interface	Category	Attempts	Success Rate	False Success	Timeout	Retry	p95 ms
`cn`	`create_adventure_campaign`	`authenticated_write`	100	0.000	100	0	0	1,305
`cn`	`like_collection`	`authenticated_write`	100	0.000	4	0	0	0
`cn`	`list_my_adventure_campaigns`	`authenticated_write`	100	0.000	100	0	0	1,333
`cn`	`list_my_characters`	`authenticated_write`	100	0.000	100	0	0	1,439
`cn`	`list_my_elementum`	`authenticated_write`	100	0.000	100	0	0	1,399
`cn`	`request_adventure_campaign`	`authenticated_write`	100	0.000	0	0	0	0
`cn`	`update_adventure_campaign`	`authenticated_write`	100	0.000	0	0	0	0
`cn`	`make_image`	`heavy_creative`	100	0.000	100	0	0	1,537
`cn`	`make_song`	`heavy_creative`	100	0.000	100	0	0	1,424
`cn`	`make_video`	`heavy_creative`	100	0.000	0	0	0	0
`cn`	`remove_background`	`heavy_creative`	100	0.000	0	0	0	0
`cn`	`get_hashtag_characters`	`public_read`	100	0.040	0	0	0	0
`cn`	`get_hashtag_collections`	`public_read`	100	0.080	0	0	0	1,404
`cn`	`get_hashtag_info`	`public_read`	100	0.040	0	0	0	0
`cn`	`list_space_topics`	`public_read`	100	0.040	0	0	0	0
`cn`	`list_spaces`	`public_read`	100	1.000	0	0	0	1,855
`cn`	`read_collection`	`public_read`	100	0.020	0	0	0	0
`cn`	`request_character_or_elementum`	`public_read`	100	0.530	0	0	0	1,572
`cn`	`request_interactive_feed`	`public_read`	100	1.000	0	0	0	2,099
`cn`	`search_character_or_elementum`	`public_read`	100	0.990	0	1	0	1,588
`cn`	`suggest_categories`	`public_read`	100	1.000	0	0	0	1,474
`cn`	`suggest_content`	`public_read`	100	1.000	0	0	0	1,922
`cn`	`suggest_keywords`	`public_read`	100	1.000	0	0	0	1,505
`cn`	`suggest_tags`	`public_read`	100	1.000	0	0	0	1,478
`cn`	`validate_tax_path`	`public_read`	100	0.990	0	0	0	1,500
`com`	`create_adventure_campaign`	`authenticated_write`	101	1.000	0	0	0	3,903
`com`	`like_collection`	`authenticated_write`	100	0.020	2	0	0	0
`com`	`list_my_adventure_campaigns`	`authenticated_write`	100	1.000	0	0	0	2,999
`com`	`list_my_characters`	`authenticated_write`	100	1.000	0	0	0	3,286
`com`	`list_my_elementum`	`authenticated_write`	100	1.000	0	0	0	3,150
`com`	`request_adventure_campaign`	`authenticated_write`	100	1.000	0	0	0	3,500
`com`	`update_adventure_campaign`	`authenticated_write`	100	0.990	1	0	0	3,798
`com`	`make_image`	`heavy_creative`	106	0.972	3	0	0	65,034
`com`	`make_song`	`heavy_creative`	109	0.972	1	2	0	110,379
`com`	`make_video`	`heavy_creative`	294	0.289	209	0	0	76,941
`com`	`remove_background`	`heavy_creative`	100	0.880	12	0	0	19,971
`com`	`get_hashtag_characters`	`public_read`	100	0.040	0	0	0	0
`com`	`get_hashtag_collections`	`public_read`	100	0.040	4	0	0	2,903
`com`	`get_hashtag_info`	`public_read`	100	0.040	0	0	0	0
`com`	`list_space_topics`	`public_read`	100	0.040	0	0	0	0
`com`	`list_spaces`	`public_read`	100	1.000	0	0	0	4,021
`com`	`read_collection`	`public_read`	103	0.000	0	103	2	40,025
`com`	`request_character_or_elementum`	`public_read`	100	0.500	50	0	0	3,732
`com`	`request_interactive_feed`	`public_read`	100	0.000	50	0	0	3,668
`com`	`search_character_or_elementum`	`public_read`	101	1.000	0	0	0	3,770
`com`	`suggest_categories`	`public_read`	100	1.000	0	0	0	4,101
`com`	`suggest_content`	`public_read`	103	1.000	0	0	0	3,129
`com`	`suggest_keywords`	`public_read`	100	1.000	0	0	0	3,256
`com`	`suggest_tags`	`public_read`	100	1.000	0	0	0	2,965
`com`	`validate_tax_path`	`public_read`	100	1.000	0	0	0	3,620

9. Skill To API Mapping

这部分补充的是 agent 真正能看到和调用的 skill 层接口，也就是这次评测实际覆盖到的 command surface。它可以清楚回答“每个接口属于哪个 skill、典型调用长什么样、文档证据在哪”。

范围说明： 这里记录的是 skill -> CLI/API 映射，不是底层后端 HTTP path 清单。如果后续需要把原始 REST / gRPC 路由也写进报告，需要继续拆 @talesofai/neta-skills 包本体，而不是只看 skill 文档。

Interface	Primary Skill	Supporting Skills	Typical Command Surface	Source Doc
`list_spaces`	`neta-space`	`neta-community`	`neta-cli list_spaces`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-space/SKILL.md`
`get_hashtag_info`	`neta-space`	`neta-community`	`neta-cli get_hashtag_info --hashtag "space_tag_name"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-space/SKILL.md`
`list_space_topics`	`neta-space`	`-`	`neta-cli list_space_topics --space_uuid "space UUID"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-space/SKILL.md`
`get_hashtag_characters`	`neta-space`	`neta-community`	`neta-cli get_hashtag_characters --hashtag "tag_name" --sort_by "hot"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-space/SKILL.md`
`get_hashtag_collections`	`neta-space`	`neta-community`	`neta-cli get_hashtag_collections --hashtag "tag_name"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-space/SKILL.md`
`read_collection`	`neta-community`	`neta-space, neta-creative`	`neta-cli read_collection --uuid "collection-uuid"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-community/SKILL.md`
`request_interactive_feed`	`neta-community`	`-`	`neta-cli request_interactive_feed --page_index 0 --page_size 10`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-community/references/interactive-feed.md`
`like_collection`	`neta-community`	`-`	`neta-cli like_collection --uuid "target collection UUID"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-community/SKILL.md`
`search_character_or_elementum`	`neta-community`	`neta-creative, neta-character, neta-elementum`	`neta-cli search_character_or_elementum --keywords "keywords" --parent_type "character" --sort_scheme "exact"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-community/SKILL.md`
`request_character_or_elementum`	`neta-community`	`neta-creative, neta-character, neta-elementum`	`neta-cli request_character_or_elementum --name "character_name"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-community/SKILL.md`
`suggest_keywords`	`neta-suggest`	`-`	`neta-cli suggest_keywords --prefix "game" --size 20`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-suggest/SKILL.md`
`suggest_tags`	`neta-suggest`	`-`	`neta-cli suggest_tags --keyword "character design" --size 15`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-suggest/SKILL.md`
`suggest_categories`	`neta-suggest`	`-`	`neta-cli suggest_categories --level 2 --parent_path "Derivative Creation"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-suggest/SKILL.md`
`validate_tax_path`	`neta-suggest`	`-`	`neta-cli validate_tax_path --tax_path "Derivative Creation>Fan Works>Honkai: Star Rail"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-suggest/SKILL.md`
`suggest_content`	`neta-suggest`	`-`	`neta-cli suggest_content --intent search --search_keywords "character,creativity" --page_size 20`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-suggest/SKILL.md`
`make_image`	`neta-creative`	`neta-character, neta-elementum`	`neta-cli make_image --prompt "@character_name, /elementum_name, ref_img-uuid, description1, description2" --aspect "3:4"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-creative/SKILL.md`
`make_video`	`neta-creative`	`-`	`neta-cli make_video --image_source "image URL" --prompt "action description" --model "model_s"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-creative/SKILL.md`
`make_song`	`neta-creative`	`-`	`neta-cli make_song --prompt "style description" --lyrics "lyrics content"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-creative/SKILL.md`
`remove_background`	`neta-creative`	`-`	`neta-cli remove_background --input_image "image_url"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-creative/SKILL.md`
`create_adventure_campaign`	`neta-adventure`	`-`	`npx -y @talesofai/neta-skills create_adventure_campaign --name "汴京最后三天" --mission_plot "..."`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-adventure/SKILL.md`
`update_adventure_campaign`	`neta-adventure`	`-`	`npx -y @talesofai/neta-skills update_adventure_campaign --campaign_uuid "campaign-uuid-here" --mission_plot_attention "..."`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-adventure/SKILL.md`
`list_my_adventure_campaigns`	`neta-adventure`	`-`	`npx -y @talesofai/neta-skills list_my_adventure_campaigns --page_index 0 --page_size 10`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-adventure/SKILL.md`
`request_adventure_campaign`	`neta-adventure`	`-`	`npx -y @talesofai/neta-skills request_adventure_campaign --campaign_uuid "campaign-uuid-here"`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-adventure/SKILL.md`
`list_my_characters`	`neta-character`	`-`	`neta-cli list_my_characters --keyword "Ada" --page_size 10`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-character/SKILL.md`
`list_my_elementum`	`neta-elementum`	`-`	`neta-cli list_my_elementum --keyword "village" --page_size 10`	`/Users/atou/Library/Mobile Documents/com~apple~CloudDocs/Neta/skills/neta-elementum/SKILL.md`

10. Artifacts

最终交付以这些文件为准：

/Users/atou/agents-in-discord/workspaces/1484560502469165306/evals/neta-skill-services/reports/20260321-comprehensive-eval-report-100x.md /Users/atou/agents-in-discord/workspaces/1484560502469165306/evals/neta-skill-services/reports/20260321-comprehensive-eval-report-100x.html /Users/atou/agents-in-discord/workspaces/1484560502469165306/evals/neta-skill-services/reports/20260321-aggregate-interface-100x.json /Users/atou/agents-in-discord/workspaces/1484560502469165306/evals/neta-skill-services/reports/20260321-aggregate-gate-check-100x.json /Users/atou/agents-in-discord/workspaces/1484560502469165306/evals/neta-skill-services/reports/20260321-aggregate-naturalness.json