Before we get into Slash, let’s set the scene.<\/span><\/p>\n Slash is the answer to all four.<\/span><\/p>\n Slash is a cloud agent that understands our codebase, our tools, our docs, and how we ship, and then does the work end to end.<\/span><\/p>\n Slash is one platform with three doors. Razorpay engineers reach it via a @Slash mention in Slack, auto-assign on a ticket, or trigger it from a GitHub CI pipeline. Same agent, same memory, same toolchain underneath. Inside, a handful of components do the actual work.<\/p>\n <\/span>Here is what happens under the hood:<\/b><\/p>\n Launch Agents: <\/b>An engineer hands Slash a task in plain English. It clones the repo, writes the code, runs tests, opens a PR, and posts an update to Slack, with no hand-holding in between. It runs in single-repo mode for the standard case, batch mode when the same change has to land across many repos at once, multi-repo mode for a task whose changes span several codebases that need to stay consistent, and clean-slate mode for skill-based work that does not need a repo at all. Execution logs stream throughout, so engineers can watch a run live or audit it later.<\/span> Slash Reviewer: <\/b>Every pull request is reviewed by a set of specialised sub-agents, each owning a single dimension: bug detection, security, code quality, the Razorpay design system, internationalization, and a pre-mortem pass. Rather than working off the diff alone, each sub-agent clones the repo and reads the surrounding file context, so its findings account for how the change interacts with the rest of the code. Severity is scored against repo-specific review skills, which encode the standards and known failure modes for that codebase, and an AI filter layer removes false positives before any comment is posted. Low-severity PRs are auto-approved without a human in the loop. About one in three code reviews flowing through Slash today is reviewed, scored, approved, and merged with no human comments.<\/span><\/p>\n Slash Event Listener<\/b>: Watches new tickets so engineers do not have to break focus mid-sprint. When a cross-functional ticket lands, it is auto-assigned to Slash, which picks it up, writes a fix that follows our standards, and posts back to the thread. In the clean case, nothing needs a human touch between ticket creation and PR approval. Scheduled Skills:<\/b> A skill is a scoped instruction that tells Slash how to handle a specific, repeatable task. A few hundred of them sit in the catalogue today, contributed by teams across the org, and we are now certifying a large chunk of them so they carry guaranteed SLAs. Typical examples include generating a sprint report, building a weekly Slack digest, analyzing open PRs for missing tests, reviewing support escalations, and running a daily Trino cost report. The routine work that used to take someone an hour every Monday morning now runs on its own. MCP and CLI Gateway: <\/b>Slash reaches tickets, GitHub, our design system (Blade), docs, logs, Grafana, deployment pipelines, the data lake, Slack, and a couple of dozen more systems through a single gateway. Each connection is registered once with scoped permissions that define exactly what Slash can read, what it can write, and what it must never touch, so access stays auditable as the platform grows. Once a tool is added at the gateway, every agent on the platform can use it without any further integration work. Discover, Slash’s Knowledge Graph: <\/b>Discover works like an internal search engine for everything the company already knows. It takes a natural language question and searches across ten plus sources, including GitHub PRs, Slack threads, Google Drive, AWS, ticketing software, and internal tooling. A leader can ask something like “what did the payments team discuss about the latency issue last week?” and get back a short synthesis along with the underlying thread, the linked tickets, and the doc someone wrote up afterwards. Making this kind of context easy to find is the first step in any culture and productivity shift, and Discover is how we are solving it.<\/span><\/p>\n Slash is the platform. Repos decide the output.<\/b><\/p>\n Connecting a coding agent to a repo takes a few seconds, but getting useful output from it can take hours, and the same agent will produce very different results across two repos. The difference comes down to how well each codebase describes itself to an agent.<\/span><\/p>\n We wanted Slash to behave with high accuracy across all our repos. So we score every repo on three pillars.<\/span><\/p>\n\n
What Slash actually does<\/b><\/h3>\n
\n<\/span><\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-launch-agents-1.gif\")
<\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-reviewer-gif-1.gif\")
<\/p>\n
\n<\/span><\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-event-listener-1.gif\")
<\/p>\n
\n<\/span><\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-skills-1.gif\")
<\/p>\n
\n<\/span><\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-mcp-gateway-1.gif\")
<\/p>\n![\"\"](\"https:\/\/blog.razorpay.in\/wp-content\/uploads\/2026\/05\/slash-discover-1.gif\")
<\/p>\n