dyad-sh/dyad
Overview
This skill reviews the current agent session for errors, snags, and hard-won knowledge, then autonomously updates project guidance files with concise, actionable learnings. It writes to topic-specific files under rules/ when appropriate, or to AGENTS.md only when a general, cross-cutting rule is needed. The skill completes without asking for confirmation and stages the changes for commit.
How this skill works
It scans the session transcript to extract up to five high-impact learnings: explicit errors, recurring gotchas, workflow friction, and architecture insights. It checks AGENTS.md and existing rules/*.md to avoid duplicates, then inserts concise 1–3 line entries into the most relevant rule file (or creates a new rules/*.md and updates AGENTS.md if needed). Finally, it runs git add to stage AGENTS.md and rules/ changes and summarizes what was modified.
When to use it
- After any debugging or build session that uncovered errors or non-obvious steps
- When a session required repeated attempts or workarounds
- If configuration, directory layout, or conventions caused friction
- At the end of a multi-step change so learnings are captured before context is lost
- When a pattern likely affects future agent runs or contributors
Best practices
- Limit additions to the five most impactful learnings from the session
- Prefer existing rules/*.md; only edit AGENTS.md for project-wide conventions
- Write entries as specific, actionable commands or recognition symptoms (include exact error text when relevant)
- Consolidate overlapping learnings into a single concise entry rather than appending duplicates
- Remove obsolete or duplicated entries in the target file before adding new ones
- Always run: git add AGENTS.md rules/ to stage changes
Example use cases
- After a failed TypeScript build, add the exact tsc error and the fix command (e.g., run npm run build before tests)
- When an Electron IPC channel caused race conditions, document the correct init order in rules/electron-ipc.md
- If environment variables were missing, add a rule pointing to the env setup and include the missing keys and their expected values
- When test flakes required a --runInBand flag, add that flag and the failing symptom to rules/e2e-testing.md
- Create a new rules/tanstack-router.md when router patterns repeatedly confuse agents or require specific imports
FAQ
No. This skill is designed to complete autonomously and stages changes without prompting for approval.
Where should I expect new learnings to appear?
Learnings go into the most relevant rules/*.md file. If no suitable rule exists and the item is general, it is added to AGENTS.md. New rule files will be created when a topic warrants it.
How many learnings will it record per session?
It records up to five concise learnings, focusing on the most actionable and impactful items.
17 skills
This skill autonomously reviews the session for errors and snags and updates AGENTS.md with actionable learnings to improve future runs.
This skill helps you rebase e2e test snapshots by extracting failed tests from PR comments and updating snapshots accordingly.
This skill adds one or more AI models to the language model constants file by researching official docs and updating specs.
This skill analyzes the past day's GitHub Actions workflow runs to detect actionable failures and open issues for remediation.
This skill analyzes session debugging data to identify errors and issues causing user-reported problems, offering actionable insights and prioritized fixes.
This skill helps you debug E2E tests by capturing targeted Playwright screenshots to visually inspect app state during failures.
This skill records a visual screencast of the PR feature using screenshots and posts it as a new comment on the PR.
This skill automatically deflakes recent E2E tests by aggregating flaky results from main CI and PRs, then proposes fixes and reports.
This skill rebases the current branch on the latest upstream changes, resolves conflicts, and pushes, saving time and ensuring a clean PR history.
This skill helps fix failing CI checks and GitHub Actions on a PR by automating lint, type, test, and build repairs.
This skill commits uncommitted changes, runs linting, fixes issues, and pushes the current branch automatically for rapid swift pushes.
This skill identifies and fixes flaky E2E tests by running them repeatedly and investigating failures to stabilize CI outcomes.
This skill reads unresolved PR comments from trusted authors and applies code changes or replies to resolve them efficiently.
This skill converts customer feedback into actionable GitHub issues, detects duplicates, and drafts follow-up reply emails to streamline issue triage.
This skill commits uncommitted changes, runs linting, fixes issues, and pushes the current feature branch to GitHub autonomously.
This skill coordinates a PM, UX, and engineering swarm to debate ideas and generate a detailed plan written to plans/$plan-name.md.
This skill coordinates three specialized sub-agents to review PR diffs, validate issues by reasoning, and post a summarized verdict.