How Lem AI builds implementation.md from a Jira branch

Most wrong implementations are built from half a ticket. The Jira description says what to ship; the Slack thread from last Tuesday says what not to break. Lem AI’s Implementation Agent merges both when you create a branch whose name matches that ticket—into a single file called implementation.md.

Why do teams lose context between Jira and code?

Engineers do not forget how to code. They forget where the decision lived. Acceptance criteria sit in Jira; the constraint about backward compatibility sits in Slack; the API shape was argued in a Google Meet; the prior attempt is in a merged PR nobody re-reads.

• Ticket descriptions written once and never updated after scope changes in chat
• Subtasks that split work without linking parent context
• Design docs in Confluence not linked from the ticket
• Meeting outcomes never copied back into Jira
• New hires who only read the ticket, not the thread

Coding agents make this worse when you prompt from memory. They start from the repo, not from the ticket graph. implementation.md is Lem AI’s way of putting the ticket graph on the branch before line one.

What triggers the Implementation Agent on git checkout?

After you install the Lem AI CLI and link a repository at getlem.ai, branch creation is the signal. Run git checkout -b with a name that embeds the ticket ID your organization uses in Jira or ClickUp.

1. Connect integrations: Slack, Jira or ClickUp, GitHub, Confluence, Meet (and Drive or Document Hub if you use them).
2. Link the repo to your Lem AI workspace.
3. Create a branch such as feature/ENG-412-billing-webhook or hotfix/lem-412-timeout.
4. Lem AI resolves ENG-412 (or your project key format) to the ticket record.
5. implementation.md is generated with cited sections from each source.

Branch naming patterns that work

• feature/PROJ-412-short-description
• fix/ENG-220-regression
• username/LEM-412-experiment (ticket ID anywhere in the name is enough)

If the branch has no recognizable ticket key, the Implementation Agent does not run—you still get search and compliance on the platform, but not automatic implementation.md for that branch.

What is inside implementation.md?

Think of it as a briefing packet for humans and LLMs—not a replacement for the ticket, but a merge of everything the ticket points at plus what Lem AI found nearby.

• Ticket title, description, status, and relevant comments
• Slack threads where the ticket was discussed or decisions were made
• Google Meet (or other) notes tied to the work
• Confluence, Drive, or Document Hub pages linked from Jira or mentioned in Slack
• Optional pointers to related GitHub PRs or files when relevant
• Inline citations so you can open the source and verify

How coding agents should use it

Open implementation.md in the same session where you edit code. Prompt once: read implementation.md, summarize the plan in bullets, then implement. You avoid re-pasting Slack into the chat and you give the model a single narrative instead of fifty disconnected files.

For token savings and fewer wrong paths, pair this file with a short prompt—the same pattern described in our article on LLM prompts and implementation.md.

Jira, ClickUp, and multi-tool stacks

Lem AI maps the ticket ID from the branch to whichever tracker you connected. Teams on ClickUp use the same workflow: branch name contains the task ID, context is gathered from connected tools, implementation.md is written for that branch.

• One workspace can connect both Jira and ClickUp if different teams use different trackers
• Slack remains the usual home for informal decisions—Lem AI links threads to tickets by references and semantics
• GitHub supplies code history; compliance features later compare diffs to ticket scope

Example workflow from ticket to merge

1. PM updates Jira ENG-412 with acceptance criteria and links a Confluence spec.
2. Engineer discusses edge cases in Slack #billing and records a Meet recap.
3. Engineer runs git checkout -b feature/ENG-412-billing-webhook.
4. Lem AI writes implementation.md; engineer reviews citations, fixes ticket if something is stale.
5. Engineer attaches implementation.md to Cursor or Claude and implements.
6. PR references ENG-412; Compliance Guard checks branch and description alignment.
7. Decision log captures any SOP answers if a new dependency was added.

The same indexed context feeds onboarding search later—a new hire can ask what ENG-412 changed and get the same sources, not a second-hand summary.

Bottom line

implementation.md from a Jira or ClickUp branch is not magic—it is assembly. Lem AI does the assembly when you branch so implementation starts from evidence. You still own the code; you stop paying the tax of rediscovering context every sprint.