You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Retrospective: Requesting the Data Dump - Multi-thread Part 1
Developer Prompt for Agent
In each thread the bug or issue was fixed or referenced, the agent prompt for this Workflow might be as follows --- be sure to replace the 1 with each sequential thread prompt:
Now that this bug is fixed, in an effort to try to gain a historical perspective on the efforts on this task, please run the attached global Workflow.
@retrospective-debug-dump-request-multi-thread.md
Note to agent: While analyzing the Agent Workflow section, replace references from __ to 1.
Notes to Developer
Important Reminder: It is imperative the developer increment the 1 in each subsequent thread prompt.
Important note: The retrospective files should be recorded in chronological order. The agent analysis is instructed to put more emphasis on the highest numbered retrospective, as it was most likely the final solution, if not the final convo.
If pasting the above into the chat prompt, press backspace at the end of the @ reference file, then select the resource from the IDE's pop-up menu. This will attach the actual resource, else, the agent likely won't find it in the Mac user's home dir (e.g., ~/.gemini/antigravity/global_workflows/*).
Agent Workflow
Step 1. Create or open a file named 'retrospective__.md' as denoted in the following directory.
File: _notes/project/retrospective__.md
Previous retrospectives should be moved after they're generated if archival is desired. If the sequenced file already exists and has data, and no override was provided in the prompt, append a _check to the filename (*_check.md), then create a new file.
Step 2. In that sequenced and empty retrospective__.md file, provide a comprehensive dump of this entire conversation thread so far. Include:
Original Goals: A summary of what we set out to achieve.
Logic & Decisioning: Explain the reasoning behind key technical decisions you made.
Unresolved Items: List any pending tasks or known bugs that still need addressing.
Full Interaction Log: A chronological list of every prompt I gave and every significant action or code change you performed.
Format this strictly as a Markdown document so another agent can easily ingest it as a context file.
Retrospective: Analysis of Data Dumps - Multi-thread Part 2
Developer Prompt for Agent
After running the requisite 'retrospective-debug-dump-request-multi-thread.md' Workflow, starting from a new agent thread, the agent prompt for this Workflow might be as follows:
To help document the 'Xyz' bug, please run the attached global Workflow. For this bug, there are __ dumps/files.
@retrospective-debugging-multi-thread.md
Reminder: Replace __ with the number of threads you had dumped.
Note to Developer
If pasting the above into the chat prompt, press backspace at the end of the @ reference file, then select the resource from the IDE's pop-up menu. This will attach the actual resource, else, the agent likely won't find it in the Mac user's home dir (e.g., ~/.gemini/antigravity/global_workflows/*).
Agent Workflow
Task: Read the retrospective__.md files noted below, and perform a critical review of the work done. Identify potential optimizations and ensure the original goals were fully met.
Multi-thread analysis
File: _notes/project/retrospective1.md
File: _notes/project/retrospective2.md
File: _notes/project/retrospective__.md
If the developer hasn't provided a count of the dumped threads, stop and ask them, as you need to know what you're working with.
Only these specific retrospective files should be used for analysis.
The agent should not consider ephemeral issues as part of the analysis. Caching or HMR types of issues are typically red herrings. If a bug is important enough to run these retro workflows on, the root cause likely had nothig to do with happenstance issues along the debugging path. Restarting a server isn't part of a root cause, it's a facepalm at best. Try to stick to actual coding or configuration solutions.
Important note: The retrospective files are in chronological order. The agent analysis should put more emphasis on the highest numbered retrospective as it was most likely the final solution, if not the final convo.
The number one goal here is to identify anything that might can help you, the agent, be able to troubleshoot and debug issues more efficiently.
Results from Task 1 should be written to:
File: _notes/project/retrospective-analysis.md
Results from Task 2 should be written to:
File: _notes/project/retrospective-lessons.md
Results from Task 3 should be appended to:
File: _notes/project/bug-diary.md
Core Retrospective Prompt
You are my engineering retrospective partner.
We just fixed a bug that could have some potential learning lessons for future bug challenges.
Your job is to help extract durable lessons and turn them into concrete rules and workflows for my future agents and projects.
As mentioned above, the #1 goal here is to see if we can extract anything from the debugging convo that might could help future debugging efforts (see Task 2).
Task 1: Debugging Review Details
Do not feel as though every bullet point needs a response. Most likely don't warrant one. In the end, we're just looking for the froth, not looking to white-paper it. Let's not over-think it.
Answer 4–8 focused questions to understand:
What the bug looked like from the user’s perspective
The real root cause (technical and process)
What slowed down detection and diagnosis
What finally led to the breakthrough
Any tooling, test gaps, or miscommunications involved
From my answers, produce three sections:
“Timeline & Key Decisions”: crisp timeline with 5–10 bullet points, highlighting decision points and wrong turns.
“What Went Well / What Hurt”: bullets for each, explicitly tying them to my answers.
“Lessons to Encode”: 5–10 concise, implementation-ready bullets phrased as habits or guardrails (e.g., “Always add a minimal failing test before touching production logic.”).
Convert “Lessons to Encode” into:
3–5 candidate Rules (always-on guardrails for future agents)
1–2 Workflows (step-by-step playbooks for debugging similar bugs)
1–3 Skills (repeatable micro-capabilities the agent should execute on demand)
For each proposed Rule/Workflow/Skill, include:
Name
When to apply it (trigger conditions)
A short, concrete description
One short example of it in action related to this bug.
Before you finalize, reflect for one short paragraph on:
Which future failures this set of Rules/Workflows/Skills is most likely to prevent
What blind spots might still remain.
Task 2: Review Analysis: Benefits
Using the analysis from Task 1, what might we learn from the referenced debugging session, if anything.
Either outline highlights of actionable sections from Task 1, provide a synopsis, or, just as valid, a simple 'nothing much to learn from here' (most times there's likely not, don't force it).
Task 3: Bug Diary & Lesson Extraction
In a readable fashion, add as many applicable items from the list below as possible to the end of the 'bug-diary.md' file, such that the app has a helpful (and searchable) bug knowledgebase of sorts:
File: _notes/project/bug-diary.md
Provide a block for:
Date fixed using yyyy-mm-dd format for consistency,
Name of Workflow file(s) used to kick off the bug agent session (if any), or if not, a short bug name,
Path(s): Provide the path to the 'retrospective' folder location as suggested in Task 4 below.
Summary: Produce a summary 5–10 line “bug diary” entry.
Then provide synopses for:
Root cause,
Wrong assumptions,
What helped, and
What revealed the truth.
Tooling imrovements: Lastly, if applicable, suggest 1–2 new or updated Rules/Workflows that could have prevented or shortened this bug. Reminder: We're looking for actionable items that can help the agent with future debugging sessions.
Provide a section separator at the end.
Reminder: The retrospective files are in chronological order. The agent analysis should put more emphasis on the highest numbered retrospective as it was most likely the final solution, if not the final convo.
Except for the first bug entry, the agent shall always use the previous entry's formatting as an example.
Task 4: Clean-up
The user should be reminded to move all 'retrospective*.md' files to a bug-specific folder after creation.
The agent should create a folder for convenience, and for use in Task 3 above.
Retrospective: Requesting the Data Dump - Single-thread Part 1
Developer Prompt for Agent
In the thread the bug was fixed, the agent prompt for this Workflow might be as follows:
Now that this bug is fixed, in an effort to try to gain a historical perspective on the efforts on this task, please run the attached global Workflow.
@retrospective-debug-dump-request.md
Note to Developer
If pasting the above into the chat prompt, press backspace at the end of the @ reference file, then select the resource from the IDE's pop-up menu. This will attach the actual resource, else, the agent likely won't find it in the Mac user's home dir (e.g., ~/.gemini/antigravity/global_workflows/*).
Agent Workflow
Step 1. Create or open a file named 'retrospective.md' as denoted in the following directory.
File: _notes/project/retrospective.md
Previous retrospectives should be moved after they're documented if archival is desired. If the file already exists and has data, and no override was provided in the prompt, rename it to retrospective_check.md (or _check1, etc.).
Step 2. In that empty retrospective.md file, provide a comprehensive dump of this entire conversation thread so far. Include:
Original Goals: A summary of what we set out to achieve.
Logic & Decisioning: Explain the reasoning behind key technical decisions you made.
Unresolved Items: List any pending tasks or known bugs that still need addressing.
Full Interaction Log: A chronological list of every prompt I gave and every significant action or code change you performed.
Format this strictly as a Markdown document so another agent can easily ingest it as a context file.
Retrospective: Analysis of Data Dumps - Single-thread Part 2
Developer Prompt for Agent
This Workflow is called from the tail end of the retrospective-debug-dump-request.md Workflow, ergo, no prompt is necessary.
Agent Workflow
Task: Read the retrospective.md file noted below, and perform a critical review of the work done. Identify potential optimizations and ensure the original goals were fully met.
Single thread analysis
File: _notes/project/retrospective.md
The number one goal here is to identify anything that might can help you, the agent, be able to troubleshoot and debug issues more efficiently.
Results from Task 1 should be written to:
File: _notes/project/retrospective-analysis.md
Results from Task 2 should be written to:
File: _notes/project/retrospective-lessons.md
Results from Task 3 should be appended to:
File: _notes/project/bug-diary.md
Core Retrospective Prompt
You are my engineering retrospective partner.
We just fixed a bug that could have some potential learning lessons for future bug challenges.
Your job is to help extract durable lessons and turn them into concrete rules and workflows for my future agents and projects.
As mentioned above, the #1 goal here is to see if we can extract anything from the debugging convo that might could help future debugging efforts (see Task 2).
Task 1: Debugging Review Details
Do not feel as though every bullet point needs a response. Most likely don't warrant one. In the end, we're just looking for the froth, not looking to white-paper it. Let's not over-think it.
Answer 4–8 focused questions to understand:
What the bug looked like from the user’s perspective
The real root cause (technical and process)
What slowed down detection and diagnosis
What finally led to the breakthrough
Any tooling, test gaps, or miscommunications involved
From my answers, produce three sections:
“Timeline & Key Decisions”: crisp timeline with 5–10 bullet points, highlighting decision points and wrong turns.
“What Went Well / What Hurt”: bullets for each, explicitly tying them to my answers.
“Lessons to Encode”: 5–10 concise, implementation-ready bullets phrased as habits or guardrails (e.g., “Always add a minimal failing test before touching production logic.”).
Convert “Lessons to Encode” into:
3–5 candidate Rules (always-on guardrails for future agents)
1–2 Workflows (step-by-step playbooks for debugging similar bugs)
1–3 Skills (repeatable micro-capabilities the agent should execute on demand)
For each proposed Rule/Workflow/Skill, include:
Name
When to apply it (trigger conditions)
A short, concrete description
One short example of it in action related to this bug.
Before you finalize, reflect for one short paragraph on:
Which future failures this set of Rules/Workflows/Skills is most likely to prevent
What blind spots might still remain.
Task 2: Review Analysis: Benefits
Using the analysis from Task 1, what might we learn from the referenced debugging session, if anything.
Either outline highlights of actionable sections from Task 1, provide a synopsis, or, just as valid, a simple 'nothing much to learn from here' (most times there's likely not, don't force it).
Task 3: Bug Diary & Lesson Extraction
Add applicable items from the list below to the end of the 'bug-diary.md' file:
File: _notes/project/bug-diary.md
Provide a section separator
Provide a short description of:
date fixed,
name of workflow file (if any), or a short bug name,
root cause, wrong assumptions, what helped, and what revealed the truth.
Produce a 5–10 line “bug diary” entry.
If applicable, suggest 1–2 new or updated Rules/Workflows that would have prevented or shortened this bug.
Task 4: Clean-up
The user should be reminded to move all 'retrospective*.md' files to a bug-specific folder after creation.
