deepdive
Full specialist analysis via parallel agent dispatch. Researcher, Architect, and PM produce a prioritized report of what to build next (30-60s).
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-14
- Stars
- ★ 1.6k
- Forks
- 491
01 前言 Front Matter
- name
- deepdive
- description
- Full specialist analysis via parallel agent dispatch. Researcher, Architect, and PM produce a prioritized report of what to build next (30-60s).
03 SKILL.md 正文
AgenTeam Deepdive
Run a full specialist analysis by dispatching three roles in parallel. Unlike the standup skill (which reads state locally), deepdive launches Codex subagents to investigate external signals, internal code health, and strategic priorities. Expect 30-60 seconds for completion.
Process
1. Auto-Init Guard
Check for .agenteam/config.yaml, .agenteam.team/config.yaml, or legacy
agenteam.yaml in the project root. If all are missing:
- Create config dir:
mkdir -p .agenteam - Copy the template:
cp <plugin-dir>/templates/agenteam.yaml.template .agenteam/config.yaml - Set the team name to the project directory name
- Generate agents:
python3 <runtime>/agenteam_rt.py generate - Tell the user: "AgenTeam auto-initialized with default roles. Edit
.agenteam/config.yamlto customize."
2. Gather State with Dispatch Flag
Call the runtime with the --dispatch flag to get state and dispatch
plans for the three specialist roles:
python3 <runtime>/agenteam_rt.py standup --dispatch
Capture the JSON output. Expected fields (same as standup, plus dispatch info):
health--on-track,at-risk,off-track, orno-active-runrun_id-- current run identifier (may be null)task-- task descriptionstages-- stage statusesartifact_paths-- map of role name to artifact directoryoutput_path-- where to write the final report (e.g.,docs/meetings/<timestamp>-deepdive.md)dispatch-- list of{role, agent}objects for the three specialist roles (researcher, architect, pm)
Create a durable checkpoint at .agenteam/deepdive/<run_id>.json before
dispatch. Record max_elapsed_minutes (default 60), max_agents (default 2
concurrent specialists), each role's attempt/thread ID, last heartbeat, output
artifact, and stop reason. On restart, validate completed artifacts and resume
only missing or interrupted roles; never repeat a completed specialist solely
because the controller restarted.
3. Dispatch Specialist Agents in Parallel
Launch three Codex subagents in parallel. Each role has a focused mandate:
Researcher (@Researcher)
Agent file: .codex/agents/researcher.toml
Prompt the researcher with:
- Read all files in
docs/research/and assess staleness (anything older than 2 weeks is potentially outdated) - Search the web and GitHub for new trends, tools, and community discussions relevant to the project
- Check for competitor moves, new releases in the dependency ecosystem, and community feedback on similar tools
- Produce a structured report of external signals
Expected output format:
## External Signals
### Trends
- [signal] description and relevance to this project
### Ecosystem
- [dependency/tool] notable updates or risks
### Community
- [source] feedback, requests, or discussions relevant to our work
Architect (@Architect)
Agent file: .codex/agents/architect.toml
Prompt the architect with:
- Read all files in
docs/designs/and compare against the current codebase -- identify design drift (where the implementation diverges from the documented design) - Check for tech debt signals: duplicated logic, overly complex modules, missing error handling, dead code
- Review dependency health: outdated packages, known vulnerabilities, abandoned upstream projects
- Produce a structured report of internal health
Expected output format:
## Internal Health
### Design Drift
- [area] how implementation differs from design doc
### Tech Debt
- [area] description and severity (low/medium/high)
### Dependencies
- [package] status and risk level
PM (@Pm)
Agent file: .codex/agents/pm.toml
Prompt the PM with:
- Wait for and read the Researcher and Architect outputs (passed as context once they complete)
- Read
docs/strategies/for current roadmap and strategic priorities - Cross-reference external signals (Researcher) with internal health (Architect) and existing strategy
- Produce a prioritized list of recommendations for what to build next, with rationale for each item
Expected output format:
## Recommendations
1. **[title]** -- rationale based on research + architecture analysis
Priority: [high/medium/low]
Effort: [small/medium/large]
2. **[title]** -- rationale
Priority: ...
Effort: ...
Dispatch order: Researcher and Architect run in parallel. PM runs after both complete (it needs their outputs as input).
4. Collect Outputs
Gather the outputs from all three subagents:
- Researcher report (external signals)
- Architect report (internal health)
- PM report (prioritized recommendations)
If any agent fails, include an error note in that section and continue with the available outputs.
Update the checkpoint and emit a heartbeat whenever a specialist reports progress. Enforce the elapsed and agent budgets. A timed-out specialist is recorded as interrupted; if it has a resumable thread, resume it before considering a fresh attempt. Synthesis is its own checkpoint and starts only after the Researcher/Architect terminal states are durable.
5. Synthesize Deepdive Report
Combine all three outputs into a single report using this format:
# AgenTeam Deepdive: <project-name>
Date: <YYYY-MM-DD HH:MM>
## Health: [ON TRACK | AT RISK | OFF TRACK]
## External Signals (Researcher)
- trending approaches, competitor moves, community feedback
## Internal Health (Architect)
- design drift, tech debt, dependency risks
## Recommendations (PM)
- prioritized list of what to build next, with rationale
## Action Items
- specific next steps with owners
Rules for the report:
- Health is derived from the runtime JSON, same as the standup skill.
- External Signals comes directly from the Researcher output. Trim to the most actionable items (no more than 5-7 bullets).
- Internal Health comes from the Architect output. Group by severity, high-severity items first.
- Recommendations comes from the PM output. Keep prioritization and effort estimates. Limit to the top 5-7 items.
- Action Items is a synthesis step you perform: extract the most concrete next steps from all three reports, assign an owner (role name) to each, and list them in priority order.
- Omit any section where the corresponding agent produced no output (e.g., if the Researcher found nothing notable, omit External Signals).
6. Write Report
Write the synthesized report to the output_path from the runtime JSON
(typically docs/meetings/<timestamp>-deepdive.md):
mkdir -p "$(dirname "$output_path")"
Write the report content to that file.
7. Display to User
Show the full deepdive report to the user in the conversation. Include a timing note:
AgenTeam Deepdive: <project-name>
Completed in ~<elapsed>s (3 specialists dispatched)
Runtime Path Resolution
Resolve the AgenTeam runtime:
- If running from the plugin directory:
./runtime/agenteam_rt.py - If installed as a Codex plugin:
<plugin-install-path>/runtime/agenteam_rt.py
Error Handling
- If a subagent fails, include a note in the corresponding section: "[role] analysis unavailable -- [error reason]"
- If the runtime command fails, fall back to a best-effort report using direct file reads (same approach as the standup skill)
- Never let one agent's failure block the entire report
Performance Target
Fast runs may finish in 30-60 seconds, but correctness does not depend on that
estimate. The durable max_elapsed_minutes budget and heartbeat/checkpoint
state govern long-running analysis. Researcher and Architect run in parallel;
PM starts after both terminal outputs are checkpointed.
04 同仓库更多 Skills
worktable-review-with-annotations
Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.
worktable-find-and-synthesize-context
Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.
worktable-create-or-update-docs
Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.
worktable-create-or-manage-records
Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.
worktable-create-or-manage-interactive-html
Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.
worktable-collaborate-threads
Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.