All notes

36 notes. Pick a domain or category, search with the search button, or open the knowledge map.

Recently updated

• 004512 load-bearing claims behind harness engineering New Harness EngineeringLectures from walkinglabs distilling the load-bearing claims about agent harnesses — why infrastructure outside the weights, not the model itself, governs how much capability gets realized.
• 0043KV cache: the engine behind fast LLM inference New LLMGenerating token N means attending to tokens 1..N-1. Recomputing their key/value tensors every step turns O(n) work into O(n²). Caching K and V is what makes real-time LLM streaming possible.
• 0035Enable LSP in Claude Code for semantic code navigation New Claude CodeClaude Code navigates code with grep by default — slow and fuzzy. Enabling the official LSP plugins gives the agent semantic goToDefinition and findReferences instead. Install, enable, restart.
• 003410 architectural patterns behind Claude Code New Claude CodeReverse-engineered from npm source maps, these 10 design decisions explain why Claude Code is fast, safe, and recoverable under production conditions.
• 0031Git submodules without surprise detached HEADsGitSubmodules embed one repo inside another by pinning a commit, not a branch. Knowing that single fact prevents most of the pain — detached HEADs, missed pushes, and silent divergence after a pull.
• 0030Branching strategies: Git Flow, GitHub Flow, trunk-basedGitHow the three dominant branching models differ in branch count, integration cadence, and release rhythm — and how to pick one based on your release model, not the model's brand.
• 0029Stacked diffs: commits as review unitsGitTreat each commit (or tiny branch) as the unit of review, not a large feature branch. Smaller reviews, less waiting, better commit hygiene — and a non-trivial rebase tax to pay.
• 0028Ignoring files beyond .gitignoreGitLocal-only ignores, untracking already-tracked files, hiding local edits to tracked files, and a global ignore for OS junk — the four problems .gitignore alone doesn't solve.
• 0027Finding the commit that broke it with git bisectGitBinary search across commit history to find the first bad commit. Works manually or fully automated with a script — turning a needle-in-a-haystack search into log2(n) steps.
• 0026Three ways to squash, and when each is rightGitInteractive rebase, soft reset, and merge --squash all collapse commits into fewer commits. They differ in granularity, on which branch they leave the result, and how much history they preserve.
• 0025Git rebase without losing workGitWhat rebase actually does to your commits, how to drive the interactive form, and the recovery moves that keep a bad rebase from becoming a lost-work incident.
• 0024Git worktrees for parallel branch workGitHow and when to use git worktree to keep multiple branches checked out at once, sharing one object store, without the disk cost or coordination headaches of multiple clones.
• 0023AWS Secrets Manager pricing modelDevOpsSecrets Manager charges two ways at once: a flat per-secret-month fee plus a per-API-call fee. Understanding the split is what lets you predict the bill and decide whether to cache.
• 0022Verifying AWS DNS records with nslookupDevOpsFour `nslookup` recipes for verifying that the CNAME, MX, and TXT records you published for CloudFront, ACM, SES DKIM, and SES MAIL FROM actually resolve before AWS does.
• 0021ACM certificate DNS validationDevOpsHow AWS Certificate Manager proves you control a domain before issuing a TLS cert: ACM publishes a `_<hash>.domain` CNAME challenge, you copy it to DNS, and renewal stays automatic forever after.
• 0020Custom MAIL FROM domain in AWS SESDevOpsWhy DMARC alignment forces you to publish a custom MAIL FROM subdomain in SES, and how to set up the MX + SPF records that prove SES can send on its behalf.
• 0019DKIM verification with AWS SES Easy DKIMDevOpsHow AWS SES proves you own a sending domain: three CNAME tokens published to your DNS host let SES sign outbound mail with DKIM on your behalf.
• 0018OAuth 2.0 authorization code flow, step by stepSystem DesignThe seven-step dance behind 'Log in with Google'. Why the redirects exist, why the code is swapped for the token server-to-server, and where PKCE plugs the public-client hole.
• 0017Passwordless authentication: OTP, SSO, and OAuth/OIDCSystem DesignWhen you stop asking users for a password, what do you verify instead? Three answers — OTP proves channel ownership, SSO and OAuth/OIDC delegate identity to a trusted issuer.
• 0016Password auth: HTTP Basic vs session cookies vs tokensSystem DesignHow a server remembers an authenticated user across HTTP's stateless requests — Basic resends credentials, session cookies pin server-side state, tokens carry signed claims. The tradeoffs that pick one.
• 0015PGP encryption: keys, signatures, and what they actually protectSystem DesignPGP gives you two independent guarantees — secrecy via encryption with the recipient's public key, and authenticity via signing with your own private key. Most production uses want both, in that order.
• 0014Hashing passwords (not encrypting them)System DesignWhy passwords are hashed, never encrypted; why fast hashes like SHA-256 are wrong for this job; and how Argon2id / bcrypt / scrypt buy you decades against an attacker with the database.
• 0013Multi-factor weighted sorting for relevanceSystem DesignCombine several signals into one comparable score by normalising each to [0,1] and taking a weighted sum. The maths is trivial; the trap is mixing raw, unbounded quantities and letting one signal dominate the rest.
• 0012Consistent hashingSystem DesignHow consistent hashing distributes keys across a changing set of nodes while keeping the number of re-mapped keys proportional to the change — and why naive modulo sharding falls apart the moment a node joins or leaves.
• 0011Scaling MySQL before shardingMySQLSharding adds an order of magnitude of complexity. Exhaust the cheaper options first — split the DB onto its own host, add a cache, add read replicas, scale the box up — and only shard when those run out.
• 0010Create a read-only MySQL userMySQLHow to create a least-privilege MySQL account for analytics or support — and the specific grants, host restrictions, and cleanup steps that keep it least-privilege over time.
• 0009ALTER TABLE without locking large tablesMySQLALGORITHM=INPLACE, LOCK=NONE lets DDL run while reads and writes continue. Works for many but not all alterations — knowing which is the difference between zero downtime and a stuck production.
• 0008OR vs IN vs BETWEEN for multi-value WHEREMySQLFor matching a column against several values, IN beats OR by orders of magnitude (binary search vs linear), and BETWEEN beats IN when the values form a contiguous range.
• 0007Reading MySQL EXPLAIN outputMySQLHow to read the EXPLAIN plan column by column — type, key, rows, filtered, Extra — what the values mean, what to fix, and how the driving table cascades through JOINs.
• 0006EXISTS vs JOIN for filteringMySQLJOIN multiplies rows then dedupes; EXISTS short-circuits at the first match. For pure filtering, EXISTS is the semantically correct choice — no DISTINCT, no optimizer roulette.