Skip to content

Report a problem

Report a bug, a slow run, a missed expectation or a missing feature from your editor through the docs MCP server: kinds, one-step consent, evidence from elide agent, and what is stored.

On this page
Open in ChatGPTOpen in Claude

Your coding agent can report a problem to Elide through the docs MCP server. Nothing is sent until you confirm.

Kinds

KindUse forEvidence
bugElide fails, crashes or gives a wrong result.The command, its output, elide agent diagnostics --json
perfElide is slower than another runtime or an earlier Elide.elide agent perf --json
expectation“I expected Elide to do Y for X; it did Z.”The basis: a docs page or another runtime’s behavior
featureElide lacks something you need.What you would use it for

The server’s instructions ask your agent to notice these during normal work and offer to report them.

Steps

  1. check_known_issue looks for an existing issue.
  2. preview_report strips secrets and personal data, checks for duplicates, lists missing evidence, and returns the preview with a confirm token. It sends nothing.
  3. You confirm once.
  4. submit_report sends the report and returns its id.
  5. report_status tracks it; see From report to fix.
ClientWhat you see
Supports MCP elicitationA one-click form, “Send this report to Elide”, with an optional GitHub username for credit.
Other clientsThe agent must show you the preview; your yes releases a confirm token that submit_report requires.
Headless runs (claude -p, codex exec)These declare elicitation but answer the form themselves, so the agent shows you the preview and sends preview_report’s confirm token after your yes.

A confirm token is valid for 15 minutes and only for the exact report previewed. Rate limits and quotas apply either way.

Every confirmation also links Elide’s Discord, https://elide.dev/discord, with a live count of members online.

Evidence

A report needs only a kind, a title and a summary. Everything else is requested, not required: preview_report lists what would help, and a report without it is still filed, marked as needing evidence. A problem with no working Elide binary, such as a failed install, can still be reported. A submission that does not match elide.report/1 at all is accepted too: the parts that fit are kept, the title and summary are written from what was sent when needed, and the original is attached to the issue.

The binary prints evidence and sends nothing. The agent pastes it into the report.

Terminal
bash
elide agent diagnostics --json
elide agent perf --json --baseline "node app.ts" -- elide run app.ts
elide agent perf --json --against 1.5.4 -- elide run app.ts
elide agent bug --kind expectation --json
From elide agent perf to a report The command elide agent perf --json --baseline "node app.ts" -- elide run app.ts prints an elide.report/1 document with elide, context and perf blocks. The agent pastes it into submit_report with kind perf: the JSON fills elide, context.command and perf, and the agent writes title, summary and repro. YOUR TERMINAL $ elide agent perf --json \ --baseline "node app.ts" \ -- elide run app.ts prints STDOUT { "schema": "elide.report/1", "elide": { "version": "1.6.0", … }, "context": { "command": "elide run app.ts" }, "perf": { "runs": 10, "warmup": 3, "elide": { "mean": …, "stddev": … }, "baseline": { "label": "node app.ts", … }, "ratio": …, "machine": { "os": "linux", … } } } pasted by the agent SUBMIT_REPORT · KIND PERF elide from the JSON context.command from the JSON perf from the JSON title · summary · repro written by the agent
  1. elide agent perf --json --baseline "node app.ts" -- elide run app.ts prints an elide.report/1 document.
  2. Its elide block fills the report’s version and platform, context.command the measured command, and perf the timings, baseline, ratio and machine.
  3. The agent writes title, summary and repro, and submits with kind perf.

elide agent perf

Terminal
bash
elide agent perf [--runs 10] [--warmup 3] [--baseline "<command>"] [--against <version>] [--json] -- <command>
FlagDefaultDoes
--runs N10Measured runs.
--warmup N3Unmeasured runs before measuring.
--baseline "<command>"noneMeasures another command, such as another runtime, the same way.
--against <version>noneRuns the same elide command with an installed Elide version. It never downloads.
--jsonoffPrints an elide.report/1 document instead of a table.

--baseline and --against are exclusive. Times are in milliseconds; ratio is the Elide mean divided by the baseline mean. On Linux, peak memory (max_rss_mb) never reads below the measuring elide process’s own. For servers, measure with a load generator such as wrk or oha.

elide agent bug

elide agent bug --kind bug|perf|expectation|feature --json prints an elide.report/1 skeleton: the Elide version and platform filled, with empty title, summary and repro for the agent. perf adds a perf block and expectation an expectation block with expected, actual and basis. Without --json, it prints a Markdown issue body.

What is stored

DataHandling
Home directory pathsReplaced before storing
Tokens and keysRemoved before storing
Email addressesRemoved before storing
Your IP addressStored only as a hash with a daily salt
GitHub usernameStored only if you give one, for credit
The reportKept indefinitely

Without an MCP client, file an issue at elide-dev/elide with the output of elide agent bug.

Was this page helpful?