Skip to content

About

Safe crash-log triage for coding agents: noisy terminal, build, test, Docker, Playwright, pytest, TypeScript, and CI failures -> compact, redacted evidence.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

17 Commits

Folders and files

Repository files navigation

FailLens Crash Gate

Crash-log triage for coding agents.

Massive crash log -> safe failure evidence -> agent continues.

FailLens Crash Gate turns noisy terminal, build, test, install, Docker, Playwright, pytest, TypeScript, local-server, and CI failures into compact redacted evidence that coding agents can safely use.

What it does

When a command fails and produces a huge noisy log, FailLens helps an agent avoid reading the whole swamp.

It returns:

  • bounded failure window
  • failure type
  • likely next action
  • redaction signal
  • safe_to_send
  • compact structured evidence

Pay-per-use xpay access

Open:

https://faillens-crash-gate.on.xpay.sh/

Generate a pay-per-use key.

Then connect the generated MCP URL to your MCP-compatible client or coding agent.

Do not commit real key-bearing MCP URLs.

Use placeholders in docs:

https://faillens-crash-gate.mcp.xpay.sh/mcp?key=YOUR_XPAY_KEY

Tools and pricing

Tool Price Use case
triage_terminal_crash $0.02 / call Normal failed terminal, build, test, install, Docker, Playwright, pytest, TypeScript, local-server, or CI logs
triage_large_terminal_crash $0.08 / call Large or very noisy failed logs

Agent rule

Give your coding agent this rule:

When a terminal command, install, build, test, Docker build, Playwright run, pytest run, TypeScript compile, local server start, or CI job fails and produces noisy logs, call FailLens Crash Gate before reasoning over the full raw log.

Use triage_terminal_crash for normal logs.

Use triage_large_terminal_crash for large or very noisy logs.

Only send the returned failure window onward automatically when safe_to_send is true.

If safe_to_send is false, stop and ask for manual review.

Hosted payload rule

Hosted xpay/MCP accepts log content.

Use:

payload
rawPayload
payloadBase64
payloadChunks
payloadBase64Chunks
log
logs
text
input
data

Do not use hosted payloadFile.

Hosted payloadFile is rejected because remote callers must not be able to make the hosted service read server-side file paths.

Rule:

Large log content: yes.
Remote server file path: no.

Invalid paid calls

xpay may charge for a tool call before the hosted service validates the payload shape.

That means invalid hosted calls, such as payloadFile, may still be charged.

For hosted usage, send log content using payload, payloadBase64, payloadChunks, or payloadBase64Chunks.

Why this saves money

A normal FailLens call costs $0.02.

A large FailLens call costs $0.08.

The expensive part is usually not the FailLens call. The expensive part is agent context waste, wrong retries, developer debugging time, repeated CI friction, and unsafe raw-log handoff.

FailLens helps turn a noisy failed command into a small, redacted evidence window that a coding agent can actually use.

See:

  • docs/COST_AND_ROI.md

Example result

Input:

npm ERR! Cannot find module @stripe/stripe-js
Process completed with exit code 1

Output shape:

decision: extract_failure
risk: clean
failure.type: missing_dependency
failure.subject: @stripe/stripe-js
safe_to_send: true
recommended_next: send_window_to_llm

What FailLens is not

FailLens is not:

  • Sentry
  • Datadog
  • Gitleaks
  • TruffleHog
  • a full debugger
  • a production observability platform
  • a generic context compressor
  • a dashboard

It does one narrow job:

failed command -> redacted bounded evidence -> agent-safe next step

GitHub Action MVP

FailLens also includes a thin GitHub Action connector for failed CI logs.

It can:

  • run a command and triage the output if it fails
  • read an existing log file
  • call the hosted FailLens route
  • write a compact redacted result to the GitHub job summary
  • avoid posting raw logs by default

Start here:

  • docs/GITHUB_ACTION.md
  • examples/github-actions/command-wrapper.yml
  • examples/github-actions/log-file.yml

Public docs

See:

  • docs/XPAY_SETUP.md

  • docs/PRIVACY_AND_LOG_SAFETY.md

  • docs/PUBLIC_PROOF_LIBRARY.md

  • docs/INVALID_CALLS_AND_BILLING.md

  • docs/XPAY_PAGE_COPY.md

  • docs/V2_6_0_LIVE_XPAY_SMOKE.md

  • docs/FAILLENS_RESCUE_PLAYBOOK.md

  • docs/FEEDBACK_TEMPLATE.md

  • docs/CASE_STUDY_TEMPLATE.md

  • docs/USER_FEEDBACK_CHECKLIST.md

  • docs/agents/README.md

  • examples/public-proof/

Value and objections

  • docs/WHY_XPAY_FIRST.md
  • docs/WHY_NOT_TAIL_GREP.md
  • docs/AGENT_BUILDER_VALUE.md
  • docs/SOLO_DEVELOPER_QUICKSTART.md
  • docs/CI_VALUE.md
  • docs/SECURITY_POSTURE.md
  • docs/BUYER_OBJECTIONS.md
  • docs/RELEASE_V2_6_2.md

Repository note

This repository is the public documentation, setup, and proof shell for FailLens Crash Gate.

The hosted FailLens service and core implementation are proprietary.

Public proof library

FailLens includes a small public proof library showing how it extracts bounded failure signals from already-public issue output.

Start here:

These examples are not endorsements from the source projects. They are public extraction examples using already-public issue details.

Usage rails

FailLens has two usage rails:

  • MCP agents use the xpay MCP route for paid per-call triage.
  • CI workflows use the GitHub Action with a protected FailLens API token.

For CI workflows, use the GitHub Action with a protected FailLens API token. The xpay MCP route remains the agent-native payment rail.

About

Safe crash-log triage for coding agents: noisy terminal, build, test, Docker, Playwright, pytest, TypeScript, and CI failures -> compact, redacted evidence.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages