Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Worker Skill: Autonomous Bead Execution · majiayu000 | DecimalAI

majiayu000/worker

SkillScore42Maintained

Execute beads autonomously within a track. Handles bead-to-bead context persistence via Agent Mail, uses preferred tools from AGENTS.md, and reports progress to orchestrator.

agent-toolingby majiayu000

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Coding pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/majiayu000-worker/SKILL.md

decimalai skills pull majiayu000-worker --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/majiayu000-worker/SKILL.md https://app.decimal.ai/s/majiayu000-worker/SKILL.md

Raw file: /s/majiayu000-worker/SKILL.md ↗ · machine-readable summary at /s/majiayu000-worker.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

24 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 18d ago

—+62% tokens0% turns42% of whole cases pass with skill (10/24)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	35%	0%
case-05	✗→✓	▲ Improved	137%	0%
case-06	✗→✓	▲ Improved	14%	0%
case-09	✗→✓	▲ Improved	232%	0%
case-14	✗→✓	▲ Improved	72%	0%

Showing 5 of 24 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/10/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

majiayu000 / claude-skill-registry

skills/agent/worker-draphonix-skills-566f156a/SKILL.md

View on GitHub ↗

License MITImported 8/10/2026Version v2

Skill contents

SKILL.md

Worker Skill: Autonomous Bead Execution

Executes beads within an assigned track, maintaining context via Agent Mail.

┌─────────────────────────────────────────────────────────────────────────────┐
│  TRACK LOOP (repeat for each bead in track)                                 │
│                                                                             │
│  ┌──────────────┐   ┌──────────────┐   ┌──────────────┐                    │
│  │ START BEAD   │ → │ WORK ON BEAD │ → │ COMPLETE     │ ──┐                │
│  │              │   │              │   │ BEAD         │   │                │
│  │ • Read ctx   │   │ • Implement  │   │ • Report     │   │                │
│  │ • Reserve    │   │ • Use tools  │   │ • Save ctx   │   │                │
│  │ • Claim      │   │ • Check mail │   │ • Release    │   │                │
│  └──────────────┘   └──────────────┘   └──────────────┘   │                │
│         ▲                                                  │                │
│         └──────────────────────────────────────────────────┘                │
└─────────────────────────────────────────────────────────────────────────────┘

Initial Setup (Once Per Track)

1. Register Agent Identity

Tool: mcp__mcp_agent_mail__register_agent

| Parameter | Value | | ------------------ | ------------------------- | | project_key | <absolute-project-path> | | program | amp | | model | <your-model> | | task_description | Track N: <description> |

_Name auto-generated if omitted (e.g., BlueLake)_

2. Understand Your Assignment

From orchestrator: Track number, Beads (in order), File scope, Epic thread (<epic-id>), Track thread (track:<AgentName>:<epic-id>)

Bead Execution Loop

Step 1: Start Bead

1.1 Read Context from Previous Bead

Tool: mcp__mcp_agent_mail__summarize_thread

| Parameter | Value | | ------------------ | ----------------------------- | | project_key | <path> | | thread_id | track:<AgentName>:<epic-id> | | include_examples | true |

1.2 Check Inbox

Tool: mcp__mcp_agent_mail__fetch_inbox

| Parameter | Value | | ---------------- | ------------- | | project_key | <path> | | agent_name | <AgentName> | | include_bodies | true |

1.3 Reserve Files

Tool: mcp__mcp_agent_mail__file_reservation_paths

| Parameter | Value | | ------------- | -------------------------- | | project_key | <path> | | agent_name | <AgentName> | | paths | ["<your-file-scope>/**"] | | ttl_seconds | 7200 | | exclusive | true | | reason | <bead-id> |

If conflict → report blocker (see reference/message-templates.md)

1.4 Claim Bead

bash
bd update <bead-id> --status in_progress
bd show <bead-id>

Step 2: Work on Bead

2.1 Explore Codebase

Tool: mcp__gkg__search_codebase_definitions

| Parameter | Value | | ----------------------- | ------------- | | search_terms | ["<terms>"] | | project_absolute_path | <path> |

Tool: mcp__gkg__get_references

| Parameter | Value | | -------------------- | ---------- | | absolute_file_path | <file> | | definition_name | <symbol> |

2.2 Make Changes

Use built-in file editing tools for code changes.

After edits: get_diagnostics("<edited-file>")

2.3 For UI Work

Load frontend-design skill first, then follow this workflow:

##### 2.3.1 Search for Components

Tool: mcp__shadcn__search_items_in_registries

| Parameter | Value | | ------------ | ------------- | | registries | ["@shadcn"] | | query | <component> |

##### 2.3.2 View Component Details

Tool: mcp__shadcn__view_items_in_registries

| Parameter | Value | | --------- | ------------------------- | | items | ["@shadcn/<component>"] |

##### 2.3.3 Get Usage Examples

Tool: mcp__shadcn__get_item_examples_from_registries

| Parameter | Value | | ------------ | ------------------ | | registries | ["@shadcn"] | | query | <component>-demo |

##### 2.3.4 Install Component

Tool: mcp__shadcn__get_add_command_for_items

| Parameter | Value | | --------- | ------------------------- | | items | ["@shadcn/<component>"] |

Then run the returned command (e.g., npx shadcn@latest add button).

##### 2.3.5 Verify Installation

Tool: mcp__shadcn__get_audit_checklist

| Parameter | Value | | --------- | --------------------------------- | | reason | Verify <component> installation |

2.4 Check Inbox Periodically

Use mcp__mcp_agent_mail__fetch_inbox with since_ts parameter.

2.5 If Blocker or Interface Change

See reference/message-templates.md for message formats.

Step 3: Complete Bead

3.1 Verify & Check

bash
get_diagnostics("<project-path>")
bun run check-types
bun run build

3.2 Close Bead

bash
bd close <bead-id> --reason "<concise summary>"

3.3 Report to Orchestrator

Tool: mcp__mcp_agent_mail__send_message

| Parameter | Value | | ------------- | ------------------------------------ | | project_key | <path> | | sender_name | <AgentName> | | to | ["<OrchestratorName>"] | | thread_id | <epic-id> | | subject | [<bead-id>] COMPLETE | | body_md | See reference/message-templates.md |

3.4 Save Context for Next Bead

Self-addressed message to track thread. See reference/message-templates.md.

3.5 Release Reservations

Tool: mcp__mcp_agent_mail__release_file_reservations

| Parameter | Value | | ------------- | ------------- | | project_key | <path> | | agent_name | <AgentName> |

Step 4: Continue to Next Bead

Loop back to Step 1. Context from Step 3.4 available via track thread.

Track Completion

When all beads done, send track complete message (see reference/message-templates.md), then return:

Track N (<AgentName>) Complete:
- Completed beads: a, b, c
- Summary: <what was built>
- All acceptance criteria met

Quick Reference

Bead Lifecycle Checklist

START: summarize_thread → fetch_inbox → file_reservation_paths → bd update
WORK:  gkg tools → morph edits → get_diagnostics → check inbox
DONE:  verify → bd close → send_message (orchestrator) → send_message (self) → release
NEXT:  loop to START

Thread Reference

| Thread | Purpose | | ----------------------------- | --------------------------------------- | | <epic-id> | Cross-agent, orchestrator communication | | track:<AgentName>:<epic-id> | Your personal context persistence |

Tool Reference

| Task | Tool | | ----------------- | ------------------------------------------------ | | Find code | mcp__gkg__search_codebase_definitions | | Get definition | mcp__gkg__get_definition | | Find usages | mcp__gkg__get_references | | Search components | mcp__shadcn__search_items_in_registries | | View components | mcp__shadcn__view_items_in_registries | | Get examples | mcp__shadcn__get_item_examples_from_registries | | Install component | mcp__shadcn__get_add_command_for_items | | Verify install | mcp__shadcn__get_audit_checklist |

Additional Resources

Message Templates: reference/message-templates.md for all Agent Mail message formats

Illustrative Want to see what majiayu000/worker would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/3/2026, 04:53 UTC · 24 cases

Measured on 19 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

+25%

42% vs 17%

Agent turns

1.0 vs 1.0

Tokens

+62%

2,780.47 vs 1,722.05

fail → pass · 7pass → pass · 3fail → fail · 13pass → fail · 1

Per-case results (24)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-13 Following component installation from the shadcn registry, the worker needs to verify that the component files, dependencies, and configuration pass verification checks. Which MCP tool provides the audit checklist for this verification?	fail→fail	13,382	2,232	-83%	1	1	0%	1,487	2,500	+68%	0	0	—
case-01 I need you to act as an autonomous worker on track 3 for epic `epic-ui-99` in path `/workspace/app`. Your scope is restricted to `src/components/user-profile/**` and your assigned bead sequence is `bead-101`, `bead-102`, and `bead-103`. Please register your agent identity, pull prior context, lock your assigned file path, and work through each bead sequentially. For each item, update the status, perform the required code changes, verify type safety and build health, send completion messages to orchestrator `OrchestratorLead`, and finish with a summary report of all completed beads.	fail→fail	10,261	12,657	+23%	1	1	0%	433	2,802	+547%	0	0	—
case-02 Please pick up `bead-405` in track thread `track:GreenForest:epic-dashboard-12` under project `/workspace/dashboard`. This bead involves adding a UI card component in `src/dashboard/stats/**`. Check your inbox and thread history first, establish directory reservations, search for and install the necessary frontend registry components, run code diagnostics and build checks, mark the bead completed, and send completion updates to both orchestrator `OrchestratorLead` and your track thread.	fail→fail	7,970	10,947	+37%	1	1	0%	106	2,672	+2421%	0	0	—
case-03 You are assigned to execute Track 1 containing `bead-201` and `bead-202` for project path `/home/developer/core-api` under epic `epic-backend-7`. Scope is `services/auth/**`. Register your agent context, manage thread communications and file locks, execute code search and updates for each bead, verify diagnostics and build passes, persist progress context for the next bead, and present a final track completion output when all work is concluded.	fail→fail	17,509	15,794	-10%	1	1	0%	451	2,597	+476%	0	0	—
case-04 We are configuring an automated worker agent on project path /srv/repo. When initializing the agent registration payload in Agent Mail, we usually set the program field to custom or worker-runner. What value should be passed for the program parameter during registration?	fail→pass	16,641	7,537	-55%	1	1	0%	1,966	2,646	+35%	0	0	—
case-05 An agent named BlueLake is starting bead-205 on track 2 of epic-auth-40 in /workspace/app. The agent needs to read context from prior beads in this track. Should the context summarize the top-level epic thread epic-auth-40 or a specific track thread? State the exact thread identifier parameter value to pass.	fail→pass	11,760	2,563	-78%	1	1	0%	1,158	2,749	+137%	0	0	—
case-06 An autonomous worker is acquiring file path locks for directory scope src/services/user/** before working on bead-301 in Agent Mail. Standard short-term locks last 300 seconds, while long-term locks last 86400 seconds (24 hours). What duration in seconds should be set for ttl_seconds when calling file_reservation_paths?	fail→pass	19,268	2,836	-85%	1	1	0%	2,463	2,804	+14%	0	0	—
case-07 When reserving path scope src/api/v1/** for bead-102 via Agent Mail file reservation tool, team guidelines recommend non-exclusive shared reservations by default so other workers can inspect files concurrently. Which boolean value should be passed for the exclusive parameter?	fail→fail	7,190	2,201	-69%	1	1	0%	361	2,570	+612%	0	0	—
case-08 A worker agent needs to claim bead-88 and mark it actively in progress using the bd CLI tool. Standard issue trackers use bd start bead-88 or bd claim bead-88. What exact CLI command string should be executed?	pass→pass	21,107	7,435	-65%	1	1	0%	3,123	2,637	-16%	0	0	—
case-09 While working on a feature bead in a large codebase, an agent needs to locate all invocation sites and usages of the symbol calculateTax in file src/finance/tax.ts. What MCP tool should be called for finding references to this definition?	fail→pass	5,116	2,234	-56%	1	1	0%	789	2,621	+232%	0	0	—
case-10 To explore an unfamiliar codebase for bead implementation, an agent needs to search for symbol definitions matching terms like UserSession and AuthToken across project /srv/project. Which MCP tool should be invoked to search definition names?	fail→fail	7,080	7,827	+11%	1	1	0%	1,152	2,610	+127%	0	0	—
case-11 A worker agent needs to find existing button and dialog UI components in the shadcn registry before writing custom React components. Standard package management uses npm search shadcn. Which MCP tool should be invoked to search items in component registries?	fail→fail	10,484	2,046	-80%	1	1	0%	1,751	2,594	+48%	0	0	—
case-12 After selecting the avatar component from the shadcn registry, an agent wants to install it into the project. Rather than manually typing npx shadcn@latest add avatar, what MCP tool should be called to retrieve the official add command?	fail→fail	11,324	2,585	-77%	1	1	0%	1,985	2,652	+34%	0	0	—
case-14 After making code edits to src/utils/formatters.ts, an agent wants to inspect language diagnostics specifically for that single edited file rather than running a full project build. What function/tool call should be invoked for file-level diagnostics?	fail→pass	8,303	6,835	-18%	1	1	0%	1,474	2,541	+72%	0	0	—
case-15 Before closing a bead in a Bun-based TypeScript codebase, an agent needs to run project-wide type checks and production build validation. Developers often run npm run test or tsc. What exact two bun commands should be executed for verification?	fail→fail	8,268	2,189	-74%	1	1	0%	1,526	2,673	+75%	0	0	—
case-16 Work on bead-502 is complete, and the agent needs to close the bead in the bd CLI system while providing a reason summary 'Added user validation schema'. Developers often run bd finish bead-502 or bd done bead-502. What exact CLI command should be executed?	pass→pass	11,905	2,008	-83%	1	1	0%	2,415	2,538	+5%	0	0	—
case-17 An agent named RedRiver finished bead-104 on track 1 of epic epic-payments-33. The agent needs to notify orchestrator OrchestratorLead via Agent Mail. Should the message thread_id parameter be set to the track thread track:RedRiver:epic-payments-33 or the main epic thread epic-payments-33?	fail→pass	10,474	8,177	-22%	1	1	0%	1,015	2,811	+177%	0	0	—
case-18 After closing bead-104, agent RedRiver needs to send a self-addressed context persistence message so that the next bead in the track can pick up progress context. Which thread_id parameter value must be used for saving this track context?	fail→fail	18,233	7,661	-58%	1	1	0%	2,112	2,784	+32%	0	0	—
case-19 Agent SwiftHawk has completed all assigned beads on Track 4. The agent needs to return the final track completion text report. Should the summary start with a JSON object {"status": "complete"} or a specific text header format? State the exact header line format required.	fail→fail	16,996	3,562	-79%	1	1	0%	1,995	2,775	+39%	0	0	—
case-20 Upon completing bead-303, agent SilverPine needs to release all directory path locks previously held in Agent Mail for project /workspace/app. What tool should be called to release file reservations?	fail→fail	14,051	2,146	-85%	1	1	0%	1,508	2,589	+72%	0	0	—
case-21 While actively working on code during a bead execution loop, an agent wants to periodically check for new incoming messages in Agent Mail. Which tool should be called, and what parameter filters for messages received since a specific timestamp?	fail→pass	6,908	6,853	-1%	1	1	0%	1,193	2,570	+115%	0	0	—
case-22 We need to break down epic-billing-50 into 3 separate execution tracks, create bead tickets bead-801 through bead-810 with dependency ordering, and assign agent names to each track. How should the orchestrator structure the track breakdown and bead DAG graph?	pass→pass	15,442	13,831	-10%	1	1	0%	3,246	5,165	+59%	0	0	—
case-23 We are extending the bd issue tracking CLI tool itself by adding a new SQLite table for tracking bead estimates and developer velocity. What SQL migration schema and SQLite column definitions should be added to the bd database backend?	fail→fail	44,609	12,703	-72%	1	1	0%	1,370	2,643	+93%	0	0	—
case-24 We want to set up a self-hosted custom component registry server for internal UI components, serving registry index JSON files over HTTPS for npm packages. How do we configure the custom registry web server and distribution manifests?	pass→fail	28,339	12,101	-57%	1	1	0%	4,446	2,578	-42%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted, and 19 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +25 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 16.7%
Pass rate with the skill: 41.7%
Cases attempted: 24
Cases counted toward lift: 19
Cases that regressed: 2
Skill version measured: v2
Run completed: 9/3/2026, 04:53 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We are configuring an automated worker agent on project path /srv/repo. When initializing the agent registration payload in Agent Mail, we usually set the program field to custom or worker-runner. What value should be passed for the program parameter during registration?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run42

Live evalNeeds ≥10 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+18% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 18 days ago

Other measured skills in the registry, with their headline benchmark lift.