Fingerprint
Andromity's Fingerprint panel shows what the coding agent read, changed, ran and delegated in a session, turn by turn, with a blast-radius view, sensitive-file checks and subagent drilldown.
Fingerprint answers "what did the agent touch?" for a session: the files it read and changed, the commands it ran, the searches it made and the subagents it started, grouped by turn. Use it to review a session before you commit, or to check whether the agent went near anything sensitive.
Opening it
- The Fingerprint button in the Waterfall panel.
- The command Fingerprint Trace & Blast Radius (
andromity.openFingerprint) from the Command Palette.
Each session has its own Fingerprint tab. It updates while the agent works and reads only your local session data.
Header
| Metric | What it counts |
|---|---|
| Agent time | Time the agent spent working on turns, not counting the time between turns |
| Files changed | Unique files written or edited, including by subagents |
| Files read | Unique files read with read_file, plus the number of searches (directory listings, file and text searches, web searches and URL fetches) |
| Commands | Commands started with shell_exec or shell_bg; reading a process's output or stopping it is not counted |
| Sensitive files | Files matching Andromity's sensitive-path rules, split into blocked (Andromity refused the access) and accessed (the access was not blocked, so the content may have reached the model) |
The header also shows Live while the agent is working and Idle otherwise.
Timeline
Each turn is a row: the turn card on the left (your prompt, its time, duration and number of actions) and its actions to the right in the order they happened. Turns run top to bottom.
To keep the view readable:
- Repeated reads or edits of the same file become one card, marked ×2, ×3 and so on.
- A turn's directory listings and file and text searches become one Explored N locations card.
- Paths are shown relative to your workspace.
- Messages you sent while a turn was running are kept with that turn, and turns from before a context compaction are kept too.
A card's badge shows its outcome: OK, Error, Blocked, Running or No result. Cards that touched a sensitive file are dashed and marked Sensitive. Read and edit cards have an Open file button.
Blast radius
The Blast radius view arranges each turn as a cluster, with the turn in the centre and everything it touched around it, so you can see at a glance how far a turn reached.
Turns and inspector
The Turns list on the left shows every turn with its prompt, time, duration and number of actions, and marks turns that hit an error or are still running. Click a turn to show only that turn.
Select a card to open the inspector on the right:
- Turn: the full prompt and any follow-up messages you sent while it ran.
- Action: each call with its status, its target or command, the tool output (long output is collapsed; select Show all) and the arguments. Sensitive files are explained here: whether Andromity blocked the access or the content may have reached the model.
- Subagent: its role, status, time and model, the task it was given, its actions and its result.
Pin the inspector to keep it open while you click around the canvas.
Subagents
Open a subagent from its card (View N actions), by double-clicking it, or from the Subagents list. The view switches to that subagent's own actions and metrics; Back to session returns.
Filters, search and navigation
- Filter by Changed, Read, Shell, Subagents or Sensitive; the matching turns stay visible for context.
- Search by file name, path or command.
- Zoom with the −, + and Fit buttons or Ctrl+scroll, and scroll to pan. The view fits the session when it opens and keeps fitting until you pan or zoom yourself.
- Drag cards to rearrange them; the layout is remembered.
- Cards and turns can be reached with the keyboard; Enter selects.
Export
Export JSON saves the whole report (turns, cards, metrics and subagents) to your workspace folder. Inside a subagent it saves that subagent's report.