Skills for database engineering in depth: PostgreSQL and MySQL internals, locks, deadlocks and vacuum, connection storms and failover, partitioning, search engines, vector indexes, ClickHouse, time series, Redis and SQLite.
Plugin databases from marketplace greenblacked-ai in greenblacked/AI. Commands read from the repo's manifest — 2 days ago.
clickhouse-parts-triageFix ClickHouse too-many-parts errors and merge backlogs by reading system.parts, system.merges and system.part_log before changing anything, then removing the cause in order of frequency: many tiny inserts (batch to tens of thousands to a million rows at about one insert per second per table, or asy
connection-storm-recoveryRecover from and prevent database connection exhaustion and reconnect herds: do the arithmetic first (instances times pool size against max_connections minus superuser_reserved_connections, and the useful ceiling of two to four active backends per core), stop the herd rather than restarting the data
deadlock-analysisTurn a database deadlock error into a code change. Collect the engine's own report, not the application stack trace: PostgreSQL's log entry naming both statements, or MySQL's LATEST DETECTED DEADLOCK section. Lay the two transactions out as a lock-order grid, classify the cycle (opposite-order batch
failover-client-hardeningMake applications survive a database failover in seconds, not minutes, by fixing the client side. Test first: force a real failover (RDS reboot with failover, Patroni switchover) under steady load, record time to first success per service and set the target below it. Then work the causes in order: p
index-pruningSafely find and drop unused, duplicate and invalid indexes to cut write cost and storage. Candidates come from pg_stat_user_indexes with idx_scan = 0 since the last stats reset, summed across the primary and every replica because statistics are per node; constraint-backing and foreign-key indexes ar
innodb-purge-lagDiagnose MySQL InnoDB undo growth and purge lag: measure the history list length (SHOW ENGINE INNODB STATUS or trx_rseg_history_len) as a trend over hours, find the oldest read view that purge cannot pass (information_schema.innodb_trx ordered by trx_started, Sleep-state threads holding an open tran
integer-key-exhaustionAudit and fix 32-bit auto-increment keys before they overflow. Measure every integer identity or serial column against its sequence (pg_sequences last_value as a share of max_value) and MySQL AUTO_INCREMENT against the column type's maximum, referencing foreign-key columns included, rank by days rem
lock-queue-triageUntangle a live blocking chain in PostgreSQL or MySQL: find the head of the chain rather than the victims by walking pg_blocking_pids() to a session that is not itself blocked, or sys.innodb_lock_waits and performance_schema.metadata_locks on MySQL, cancel a waiting ALTER first because it queues eve
postgres-full-text-searchBuild search inside PostgreSQL and know when to stop: decide with numbers first (corpus size, query rate, faceting, synonyms), then add a stored generated tsvector column (PostgreSQL 12 and later) with setweight, a GIN index, websearch_to_tsquery for raw user input instead of to_tsquery, ts_rank_cd
redis-memory-triageWork out why Redis is full, evicting or out of memory: decide whether the instance is a cache, a durable store or a queue (they need different maxmemory-policy values and separate instances), read INFO memory (used_memory against used_memory_rss, overhead, client output buffers, replication backlog)
search-mapping-designDesign Elasticsearch or OpenSearch mappings that stay small, fast and changeable: start from the queries, aggregations and sorts the index must serve, choose text for analysed matching and keyword for exact filters, aggregations and sorts, set dynamic to strict in the index template with dynamic_tem
search-relevance-tuningImprove Elasticsearch or OpenSearch ranking against measured judgments instead of anecdotes: build a graded judgment set first (top queries by volume, zero-result queries, the complaint list, graded 0 to 3 by someone who knows the product, kept in version control), baseline with _rank_eval (NDCG@10,
shard-allocation-triageResolve red or yellow Elasticsearch and OpenSearch clusters and unassigned shards in a fixed reading order: _cluster/health, _cat/shards with unassigned.reason, then _cluster/allocation/explain read decider by decider. Causes in frequency order: disk watermarks (flood stage sets read_only_allow_dele
sqlite-in-productionRun SQLite reliably as the database of a server or desktop application: set the per-connection pragmas (journal_mode WAL, synchronous NORMAL, busy_timeout, foreign_keys ON, PRAGMA optimize before close), keep one writer and a pool of readers, begin every writing transaction with BEGIN IMMEDIATE beca
table-partitioningDecide whether to partition a large PostgreSQL or MySQL table, and do it without an outage. Gate on a named reason (retention by dropping a partition, pruning on a column in nearly every query, or per-partition maintenance); a table that fits in memory with no retention need stays whole. Verify prun
timeseries-retention-downsamplingDesign retention, rollups and compression for a time-series store: compute ingest rate, bytes per point and active series cardinality first, write a resolution-versus-retention table driven by what old data is asked, build the rollup (TimescaleDB continuous aggregate, ClickHouse materialized view in
typeahead-search-designBuild search-as-you-type and autocomplete on Elasticsearch or OpenSearch: pick the mechanism by content (completion suggester for prefixes of curated terms, search_as_you_type with bool_prefix for mid-phrase matching on documents, edge_ngram at index time with a plain standard analyser at search tim
vacuum-bloat-triageDiagnose and fix PostgreSQL table and index bloat when dead tuples keep accumulating: measure first with pgstattuple_approx and pg_stat_user_tables (n_dead_tup, last_autovacuum, autovacuum_count), then find what pins the xmin horizon (oldest backend_xmin in pg_stat_activity, replication slots, prepa
vector-index-tuningTune an approximate nearest-neighbour index against measured recall and latency: compute exact top-k by brute force for about 1,000 sampled real queries, score recall@k as overlap with the index's answers, tune to a stated target such as recall 0.95 at k=10 and a p95 rather than defaults. Covers pgv
xid-wraparound-defenseMonitor and recover from PostgreSQL transaction ID and multixact wraparound risk: read age(datfrozenxid) per database and age(relfrozenxid) and mxid_age(relminmxid) per table oldest first, alert near 500 million of the roughly 2.1 billion ceiling so a page arrives with weeks of runway, and in an eme
deadlock-log-readerRead the deadlock reports in a PostgreSQL server log or a MySQL error log (innodb_print_all_deadlocks output) and return the cycle class of each distinct pattern, with counts and the log line that decided it, so the log itself never enters the caller's context. Use when a log file holds many deadloc
Skills for making games: build a playable prototype and take it to a finished game, hold a frame budget on the hardware you are shipping to, choose a multiplayer authority model and make it survive latency, tune numbers against telemetry instead of taste, and keep an asset pipeline from eating the build.
39 skills · 2 commands · 2 agents
Skills for the defensive side of shipping software: review a change for the vulnerability classes that actually get exploited, reduce who can do what toward least privilege without breaking production, rotate or contain a credential, harden a container image and keep checking it, read a Terraform plan for what it destroys, preserve evidence and build an intrusion timeline after a compromise, respond to credential stuffing, write and test detection rules, prove tenant isolation, handle vulnerability reports from outside researchers, and reverse engineer the binaries, managed CIL assemblies, APK packages, packaged apps, firmware and WebAssembly you are authorised to examine — including de-bloating a smart-TV or streaming device you own.
31 skills · 2 commands · 5 agents
Skills for the daily loop of changing code: review a diff for the defects that actually break production, debug a failure down to a proven cause, decide what to test before writing tests, restructure code without changing behaviour, orient yourself in an unfamiliar codebase, review a regex for catastrophic backtracking, write documentation that stays true, scaffold new production-ready code, design an API you can still evolve, land a dependency upgrade, build a website, and author a skill that fires when it should.
29 skills · 1 command · 5 agents
Skills for keeping a running system alive: triage a red pipeline to its real cause, get a broken Kubernetes workload back before diagnosing it, design alerts where every page is actionable, instrument a service so the next incident is answerable, write the runbook a stranger can follow at 3am, rehearse failure on purpose, plan for load before it arrives, and find out why the bill grew.
19 skills · 2 commands · 7 agents
Skills for engineering leadership: take over a team and audit what you inherited, record decisions, write status updates, run postmortems and incident comms, read delivery health, review on-call load, design the team cadence, a reorganisation and goals, make the case for headcount or for paying down tech debt, map team skills and single points of knowledge, plan onboarding, hire, benchmark pay, evaluate vendors, handle retention, growth reviews, improvement plans and difficult conversations, and assess how a team uses AI. Also ships two subagents, for reading contracts and synthesising feedback, and two commands.
22 skills · 2 commands · 2 agents
Skills for code that runs on a user's browser or phone: field performance, hydration, memory, version skew, service workers, CSP and Safari bugs, dates and locales, crashes, hangs, startup, battery, bad networks, on-device migrations, deep links, pushes and app size.
22 skills · 1 agent
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
15 skills · hooks
Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications
5 skills · hooks
Up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your coding agent's context.
1 MCP server
Build powerful, stateful CLI interfaces for any GUI application using the cli-anything harness methodology.
5 commands
Automates browser interactions for web testing, form filling, screenshots, and data extraction
12 skills
Perform AI code review on Git diffs — supports workspace changes, branch ranges, and single commits with concurrent per-file analysis, codebase search, and deep context-aware review.
2 commands
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.