53 skills, sorted by phase.
Every name starts with ab-, which keeps them apart from other skill packs in a shared skills folder. Search by name or summary, or pick a phase.
Skills
53 of 53 skills
- Pipelines
ab-build-pipeline
Runs a feature through eight supervised stages (discuss, brainstorm, plan, execute, review, verify, optional deploy check, knowledge capture) with a checkpoint after each
Use when: A feature you approve stage by stage
- Pipelines
ab-ship-pipeline
Ships a feature end to end with no checkpoints, tracked in
.agent-blueprint/run/state.json; the ship runner drives it unattendedUse when: A well-defined feature you want built hands-off
- Pipelines
ab-quick-fix
A small, well-understood change through a short test-first loop, then a commit on a branch
Use when: Under three files, obvious approach
- Pipelines
ab-orchestrate
Runs a plan as team work: ledger, dependency-ordered waves, file ownership or worktrees, lead-only commits, review and sign-off
Use when: Four or more tasks, some independent
- Design
ab-brainstorming
Turns an idea into an approved design: settles what the repository answers, challenges the premise, compares two or three approaches, saves the plan. Never asks twice for a design you approved
Use when: Before any new feature
- Design
ab-discuss
Captures and locks your decisions before planning, with prior art from ADRs, plans and learnings
Use when: Ambiguous choices ahead of a plan
- Design
ab-ideation
Generates grounded improvement ideas from learnings, structure and git history, and saves five to seven ranked survivors
Use when: "What's worth building?"
- Design
ab-writing-plans
Turns an approved design into a plan that records decisions, not code: file paths, each test and what it asserts, signatures, verification commands, review focus
Use when: After design approval
- Design
ab-deepen-plan
Enriches a plan with five read-only research helpers in parallel; findings go under each section as research notes
Use when: A plan that needs more evidence
- Design
ab-deep-research
Researches a topic with five helpers in parallel and writes one synthesized brief
Use when: Before planning something unfamiliar
- Design
ab-spike-exploration
A timeboxed, hands-on spike with throwaway code that answers one technical question, reported to
docs/research/Use when: Significant technical uncertainty
- Design
ab-scope-cutting
Cuts a feature to the smallest useful deliverable with MoSCoW and checks the must-haves still ship
Use when: Feature too large or deadline at risk
- Execution
ab-executing-plans
Executes a plan in batches of three tasks with a checkpoint after each batch and a whole-branch review at the end
Use when: Plan execution in this session
- Execution
ab-subagent-driven-development
A fresh helper per task, then a spec reviewer and a code reviewer, with fix rounds until both pass
Use when: Plan execution with two-stage review
- Execution
ab-autonomous-loop
Runs a plan's tasks with no checkpoints: verify, tick, commit, retry after a written reflection, stop on a circuit breaker or a risk score
Use when: "Just do it all"
- Execution
ab-test-driven-development
Red, green, refactor: one failing test, the least code that passes, the whole suite, then refactor. Keeps code that existed before the task
Use when: Any new code
- Execution
ab-source-driven-development
Writes framework- and library-specific code from the official docs for the installed version, citing the URL or marking it unverified
Use when: Framework or library APIs
- Execution
ab-dispatching-parallel-agents
One focused helper per independent problem, all started at once, then integrated
Use when: Two or more unrelated failures
- Execution
ab-using-git-worktrees
An isolated git worktree on a new branch for feature or parallel work
Use when: Before major features
- Quality
ab-systematic-debugging
Root cause before any fix: classify, reproduce, trace, one hypothesis at a time against evidence tiers, fix test-first
Use when: Any bug or test failure
- Quality
ab-verification-before-completion
Fresh evidence before any completion claim: the output that would prove it false, the full command, its exit code
Use when: Before saying work is done
- Quality
ab-requesting-code-review
A fast single-reviewer review of a guarded range with the code-reviewer helper
Use when: After a task or before a merge
- Quality
ab-receiving-code-review
Acts on review feedback by verifying each item first, then one change at a time with a test, or a technical pushback
Use when: When review feedback arrives
- Quality
ab-review-swarm
Specialized reviewers in parallel, findings validated and merged into one P1/P2/P3 report
Use when: Significant changes
- Quality
ab-iterative-refinement
Review, fix and review again until the findings converge (fast, deep or perfect)
Use when: Ship pipeline reviews,
ab-build-pipeline --iterate N - Quality
ab-add-tests
Backfills tests for code that has none, ranked by risk, with you choosing which gaps to fill
Use when: Improving coverage
- Quality
ab-browser-testing
Verifies UI changes in a real browser through a browser automation tool, including error states and viewports
Use when: After UI changes
- Quality
ab-document-review
Three-pass review of a document (accuracy against the repository, clarity, completeness) with a verdict
Use when: Specs, plans, READMEs, runbooks
- Completion
ab-finishing-a-development-branch
Tests, a plan audit against the diff, then merge locally, push and open a PR, or keep the branch; discards only on request
Use when: After all tests pass
- Completion
ab-pr-workflow
The pull request lifecycle: checks on the pushed commit, a body that states the motivation and is scanned for secrets, self-review, comment resolution, merge onto a green main
Use when: Creating or finishing a PR
- Completion
ab-session-wrap
Ends a session from git history and the file system: rewrites the Session Continuity notes in
docs/context/STATUS.md, records learnings, updates the docsUse when: End of a session
- Completion
ab-knowledge-compounding
Records a solved problem in
docs/solutions/with search terms and cross-linksUse when: After a non-trivial problem
- Completion
ab-changelog-generation
Release notes from git history in Keep a Changelog format
Use when: Preparing a release
- Session management
ab-project-start
Scaffolds
AGENTS.md,CLAUDE.md,docs/context/andBACKLOG.mdby merging, then fills in conventions, goals and statusUse when: New project, or adding the blueprint to one
- Session management
ab-project-status
Where the project stands: code state, work in flight, goal progress, top three next actions
Use when: Start of a session
- Session management
ab-resume-session
Reloads what earlier sessions left and checks whether HEAD moved since the handoff
Use when: Picking up earlier work
- Session management
ab-pause-checkpoint
A quick mid-session snapshot plus the execution state in
docs/context/STATE.mdUse when: Stepping away
- Session management
ab-context-checkpoint
A timestamped recovery point with progress, decisions, next steps and open questions
Use when: Before risky operations or a long session
- Session management
ab-session-continuity
Keeps
docs/context/STATE.mdcurrent across session boundaries with a HEAD stampUse when: During wave orchestration
- Session management
ab-backlog-triage
Sorts the
BACKLOG.mdinbox into prioritized, typed items against the goalsUse when: When the inbox grows
- Session management
ab-health-check
Eight areas (build, tests, lint, dependencies, conventions, docs, backlog, git) checked with the project's own commands
Use when: Periodic check
- Operations
ab-codebase-mapping
Maps an unfamiliar codebase, read-only, into structured documentation with file paths
Use when: Before modifying unfamiliar code
- Operations
ab-resolve-in-parallel
Fixes a batch of independent items at once, one helper per item, then checks for conflicts
Use when: PR comments, review findings, test failures
- Operations
ab-deployment-verification
A go/no-go check across eight areas with concrete evidence
Use when: Before a production deployment
- Operations
ab-migration-planning
A migration plan where every step can be undone: blast radius, expand-contract steps, rollbacks
Use when: Database, API or dependency migrations
- Operations
ab-performance-profiling
Measure, profile, one optimization at a time against the noise floor, reverted attempts recorded
Use when: When something is slow
- Operations
ab-dependency-management
Adds, upgrades and removes dependencies through five gates, pinned and committed with the lockfile
Use when: Dependency changes
- Operations
ab-forensics
Diagnoses a failed, stalled or aborted automated run after the fact from its logs, run state and git history
Use when: A ship run that did not finish
- Orchestration primitives
ab-wave-orchestration
Groups dependent tasks into waves of parallel helpers with an integration check between waves
Use when: Four or more tasks with mixed dependencies
- Orchestration primitives
ab-swarm-orchestration
Several read-only specialist helpers on the same input, one dimension each, merged by a synthesizer
Use when: Custom swarms
- Meta
ab-writing-skills
Writes, edits and tests skills that load in every supported tool: agentskills frontmatter, the 8,000-byte cap, capability snippets, prompt files, a test first
Use when: Creating or changing skills
- Meta
ab-plugin-update
Upgrades Agent Blueprint itself through each tool's own updater, or a new run of the installer for Hermes and Amp, then checks the version. Manual-only
Use when: Upgrading the blueprint
- Meta
ab-migrate
Cleans a v3 install out of a project and renames
CLAUDE.mdtoAGENTS.mdbehind a backup branch. Manual-onlyUse when: Moving a project from v3
No skill matches “”.
Clear the search and filterHelper prompts
30 helper prompts live inside the skills that use them. Where the tool has subagents, a security review or a plan check runs in its own context, with only the inputs its prompt file names. Elsewhere the main session follows the same file.
ab-deepen-plan
Design1 promptplan-checkerPlan validation, gap detectionWhen it runs: After writing a plan, before executionAlso used byab-build-pipelineandab-ship-pipeline
ab-deep-research
Design6 promptsbest-practices-researcherIndustry standards, library documentationWhen it runs: When external guidance is neededAlso used byab-deepen-plangit-history-analyzerHow the code evolved and whyWhen it runs: When the reason behind code mattersAlso used byab-deepen-planandab-ideationlearnings-researcherPast solutions, decisions, patternsWhen it runs: Before planning: searchesdocs/for prior artAlso used byab-build-pipeline,ab-deepen-plan,ab-ideationandab-ship-pipelineresearch-synthesizerCombining the research helpers' outputWhen it runs: After parallel researchframework-docs-researcherCurrent framework docs for the installed versionsWhen it runs: Before planning features that use specific framework APIsAlso used byab-build-pipeline,ab-deepen-planandab-ship-pipelinecodebase-context-mapperThe files and dependencies one change touchesWhen it runs: Before planning a changeAlso used byab-build-pipeline,ab-deepen-plan,ab-ideationandab-ship-pipeline
ab-executing-plans
Execution1 promptpattern-mapperExisting files that new code should followWhen it runs: Between research and execution
ab-subagent-driven-development
Execution2 promptsimplementerOne plan task, test first, with a self-reviewWhen it runs: Subagent-driven development: a fresh helper per taskspec-reviewerBuilt what the task asked, nothing moreWhen it runs: Subagent-driven development: after each implementer, before code review
ab-systematic-debugging
Quality1 promptbug-reproduction-validatorBug reproduction, fix verificationWhen it runs: When debugging: validates repro steps and fixes
ab-review-swarm
Quality12 promptscode-reviewerStandards, correctness, plan complianceWhen it runs: After a major step or before mergeAlso used byab-finishing-a-development-branch,ab-requesting-code-review,ab-subagent-driven-developmentandab-wave-orchestrationarchitecture-strategistStructural patterns, service boundariesWhen it runs: Reviewing PRs, adding services, refactoringsecurity-sentinelOWASP, auth flows, vulnerability scanningWhen it runs: Before deployment, after auth, payment or API workcode-simplicity-reviewerUnneeded code and over-engineeringWhen it runs: After implementation is completeperformance-oracleBottlenecks, N+1 queries, algorithmic complexityWhen it runs: After features are built, on performance concernsschema-drift-detectorUnrelated schema or migration changesWhen it runs: Reviewing PRs: catches scope creep in the data layerfrontend-reviewerUI code qualityWhen it runs: Reviewing frontend code: accessibility, responsiveness, performanceconvention-enforcerCONVENTIONS.mdcomplianceWhen it runs: Reviewing code against project standardsdata-integrity-guardianMigration safety, transactions, rollback plansWhen it runs: PRs with migrations, schema changes, data transformstest-coverage-reviewerTest quality, meaningful assertions, edge casesWhen it runs: After implementation: checks that tests validate behaviorfindings-synthesizerMerging review swarm outputWhen it runs: Afterab-review-swarm: removes duplicates and prioritizes the findingsfindings-validatorIndependent re-check of review findingsWhen it runs: Betweenab-review-swarmand synthesis: drops false positives
ab-add-tests
Quality1 prompttest-gap-analyzerCoverage gaps, test generationWhen it runs: Improving coverage or before major refactors
ab-document-review
Quality1 promptdoc-claim-verifierClaims in docs against the codebaseWhen it runs: Reviewing READMEs, ADRs, runbooks after a refactor
ab-pr-workflow
Completion1 promptpr-comment-resolverTargeted PR comment resolutionWhen it runs: Processing review feedback: one comment per helperAlso used byab-resolve-in-parallel
ab-codebase-mapping
Operations1 promptcodebase-mapperArchitecture, conventions, stack analysisWhen it runs: Onboarding to unfamiliar code or before modifying it
ab-deployment-verification
Operations1 promptdeployment-verifierDeployment readinessWhen it runs: Before deploying: checks eight areasAlso used byab-ship-pipeline
ab-wave-orchestration
Orchestration primitives1 promptintegration-verifierCross-task integrationWhen it runs: After a wave: checks that parallel work fits togetherAlso used byab-orchestrate
ab-swarm-orchestration
Orchestration primitives1 promptintegration-checkerComponent wiring, connection validationWhen it runs: After implementation: checks that components connect