Skills

skills are the way you teach resolve a specific capability — a triage flow, a rollback procedure, a postmortem template, a standard report each skill is a small, self contained set of instructions that resolve loads on demand when the task calls for it skills follow the open agent skills https //agentskills io/home standard, the same format used by claude and a growing list of agentic tools a skill you've already written for another agent will work in resolve, and a skill you write in resolve travels with you why skills? resolve already reads your team knowledge — your resolve md, runbooks, and docs — to understand your environment that works well for interpretive knowledge conventions, glossaries, system overviews, things the agent should reason about what knowledge as prose handles less reliably are procedures the exact rollback sequence with the right flags, the five query triage flow your senior on call runs on every payments alert, the postmortem template you've spent a year refining when procedures live as prose, the agent has to interpret them, and interpretation drifts — a paraphrased step here, a flag re derived there skills package the procedural parts of your operations so the agent invokes them directly instead of paraphrasing them out of prose a wiki page about your canary rollback tells the agent what to do a skill is the rollback skills are especially useful when a task is performed frequently the investigation steps are well defined the output should be consistent across runs the action depends on team specific tooling, dashboards, or conventions how skills work resolve loads skills using progressive disclosure , so you can register many skills without bloating the agent's context discovery resolve loads only the name and description of every available skill at the start of a task — just enough to know when each might be relevant activation when a task matches a skill's description, resolve pulls the full instructions into context execution resolve follows the instructions in the skill body this applies everywhere resolve reasons about your systems alert investigations automatically pick up skills whose descriptions match the alert the assistant (chats and tasks) picks up skills whose descriptions match the user's request slash commands let users invoke a skill directly by name (see slash commands docid\ utdczovphczhps0bjyc7y below) what a skill looks like a skill is a markdown file with yaml frontmatter the minimum is a name and a description \ name payments db connection pool description use when investigating latency or error rate spikes on the payments api service, especially when the alert references database wait times or 5xx errors from /charge or /refund endpoints \ the payments api service runs on a shared connection pool with 50 max connections sustained db pool active above 90% for 60+ seconds causes request queueing and downstream 5xx errors before the database itself shows distress when investigating latency on payments api, check db pool active first — it leads the latency curve by 30 seconds the pool size is intentionally tight to push backpressure upstream rather than fan out load to the db; do not recommend increasing pool size as a fix without consulting the payments team the frontmatter is what resolve sees during discovery the body is loaded only when the skill activates resolve currently supports skill md only the agent skills specification also allows skills to bundle scripts, reference files, and assets in adjacent directories resolve will load the skill md body, but bundled scripts and file references aren't executed or fetched yet support is on the roadmap — until then, keep all skill content inline in skill md frontmatter fields field required notes name yes 1–64 characters lowercase letters, numbers, and hyphens only no leading/trailing or consecutive hyphens description yes up to 1024 characters should describe what the skill does and when to use it the clearer the trigger, the more reliably resolve picks it up resolve supports the optional fields defined in the agent skills specification https //agentskills io/specification ( license , compatibility , metadata ) most skills don't need them creating skills in resolve skills live on the skills section in your team page (teams > yourteam) there are 3 ways to add them 1\ start from a resolve template click + add > import from template to pick from resolve provided templates such as deployment monitoring, common triage flows, postmortem generation, and more! this is the fastest way to get started and you can customize it for your environment 2\ import from external sources click + add > import from github to connect a git source resolve auto detects any folder following the open agent skills format and shows you the list — you can filter which skills to import, so the same repo can serve multiple agents the latest version is fetched at activation time, so updating the source updates the skill in resolve 3\ write a skill from scratch click + add > new skill and write the skill md content directly in the in app editor useful when you're encoding a procedure you've never written down before once a skill is in the system, it's immediately available to alert investigations, chats for your team, and as a slash command — no further configuration required slash commands slash commands are how users invoke a skill directly, instead of leaving it to resolve's discretion in any assistant input — chat or task — type / to bring up an autocompleted list of available skills, then pick one slash commands are useful when you want a deterministic invocation you know exactly which skill should run, and you don't want the agent deciding for itself common patterns standard reports — end of shift handoffs, weekly reliability summaries, deploy reports opinionated triage flows — the multi step investigation your senior on call runs from muscle memory templates — postmortems, status page updates, incident summaries in your team's voice focused lookups — service ownership, recent deploys, the right runbook for an alert any skill can be invoked as a slash command the command name is the skill's name field, and the prompt after the command is passed through as input example an /oncall handoff skill a report style skill that produces consistent end of shift output, so the next on call gets the same view every time \ name oncall handoff description generate the end of shift on call handoff use when the user types /oncall handoff or asks for a shift summary, handoff notes, or what's still open from the last shift \ produce a handoff covering the last 12 hours 1\ open incidents — anything still in an active state, with the latest update timestamp and the current owner 2\ pages fired — alerts that paged during the shift, grouped by service, with status (acknowledged / resolved / still firing) 3\ still smoldering — services with elevated error rates or saturation that didn't quite cross alert thresholds but are worth watching 4\ deploys in flight — anything mid rollout or paused \### output \ use the four headings above, even if a section is empty (write "none") \ keep each bullet to one line; link to the incident or alert when possible \ end with a one sentence "top thing to watch" recommendation example a /triage payments latency skill an opinionated triage flow that encodes the order your senior on call investigates a known alert class \ name triage payments latency description run the standard triage flow for latency alerts on the payments api service use when investigating /charge or /refund p95 latency spikes, payments api slo burn alerts, or when the user asks to triage payments latency \ run these checks in order and stop as soon as you find the cause 1\ db connection pool — check `db pool active` on the `payments overview` grafana dashboard sustained > 90% for 60+ seconds is the most common cause of payments latency and leads the latency curve by 30 seconds 2\ recent deploys — list payments api deploys in the last 2 hours correlate deploy timestamps against the latency spike 3\ upstream dependencies — check the auth service and ledger service dashboards for their own latency or error rate elevation payments latency often inherits from upstream 4\ queue depth — check `payments outbox lag` on the same dashboard sustained growth indicates a downstream consumer is behind 5\ logs — only if 1–4 are clean, search payments api error logs for the spike window \### output \ lead with which step found the cause (or "no clear cause from steps 1–5") \ include the specific metric value or log signature that confirmed it \ do not recommend increasing the connection pool size as a fix — the tight pool is intentional page the payments team instead best practices the full guidance lives at agentskills io/skill creation/best practices https //agentskills io/skill creation/best practices a few highlights worth keeping in mind for resolve write descriptions that describe when to use the skill, not just what it does the description is the only thing resolve sees during discovery, so vague descriptions activate at the wrong times — or not at all include the keywords, alert names, services, or phrases that should trigger the skill keep skills focused a skill that triages payments latency and generates a postmortem and runs a rollback is too broad split it skills compose well; over stuffed skills don't be specific about data sources and output format "check the dashboard" is weaker than "query the payments overview grafana dashboard, panel db pool active , and report the 5 minute average " specificity is what makes skills repeatable add what the agent wouldn't know skip the explanations of http, sql, or your observability tool focus on the project specific conventions, gotchas, and judgment calls only your team knows iterate from real use when resolve runs a skill and you correct it, fold the correction back into the skill the skill heatmap on the skills page shows how often each skill was invoked and for what alerts — use it to tune descriptions and catch skills that aren't being picked up how skills relate to other knowledge skills sit alongside the rest of team knowledge docid wyms9kn coq eyvqdehf each type plays a different role knowledge type best for resolve md general context, conventions, glossaries, system overviews — things the agent should reason about alert runbooks step by step guidance pulled in deterministically when a specific alert fires dashboard guidance how to read and interpret a specific dashboard's panels docs general reference material — architecture diagrams, design docs, on call rotations skills specific, repeatable procedures invoked dynamically based on the task — including everything that used to be a slash command skills don't replace your other knowledge they give the procedural parts of your operations a format that lets them run exactly the same way every time alert runbooks can even reference skills directly "for latency alerts on payments api, invoke the payments db connection pool skill before checking application logs " further reading agent skills overview https //agentskills io/home — the open standard resolve implements agent skills specification https //agentskills io/specification — full format reference best practices for skill creators https //agentskills io/skill creation/best practices — writing skills that activate reliably and stay focused optimizing skill descriptions https //agentskills io/skill creation/optimizing descriptions — tuning the description field so skills trigger on the right prompts