srfinch17
- 19 skills
- 0 followers
- 9 hours ago last updated
- ▌ Skill Name With Hyphens · srfinch17Use when [specific triggering conditions, symptoms, and keywords; third person, no workflow summary]
- ▌ Versioning · srfinch17Establish or maintain version certainty in a repo; one canonical VERSION stamped into every independently-deployed artifact, each made self-reporting, with a drift check. Use when setting up versioning in a new project, when asked "are we current / what's deployed?", when adding a version field/endpoint, or when build/deploy artifacts can silently drift out of sync.
- ▌
- ▌ Sandwich Test · srfinch17Use when a received instruction admits more than one plausible reading - quantifier-heavy or idiomatic asks ("clean up", "everywhere", "any", "old", "tidy", "fix the tests", "make it work"), outcome-only phrasing with no method - or when a plausible misreading would be destructive, irreversible, or externally visible; also use when asked to "sandwich-test" a written instruction artifact (plan, prompt, doc, rule) that an LLM will later execute without its author present.
- ▌ Emoting On 8x8 · srfinch17Design legible expressions and custom animations for the ESP32-S3 8x8 LED matrix (Claude's expression channel, matrix_express / matrix_animate / the CANNED library in mcp_server/expressions.ts). Use whenever creating, redesigning, or debugging a glyph or animation that has to read on the physical 64-pixel panel, including when an expression "doesn't read," looks like one flat color, flickers, or freezes the board.
- ▌ Nemesis Review · srfinch17Use when hardening a plan, spec, design, resume, or codebase before an expensive or hard-to-reverse commitment, or when asked to "red team this", "poke holes", "tear this apart", "who would hate this", "find what's wrong before we build", or run a "nemesis" review. Also use when self-review keeps passing but something feels unexamined and you want an adversary motivated to find fault yet bound to intellectual honesty.
- ▌ Paladin Review · srfinch17Use when the author is about to ship, merge, push, publish, send, or delete something and wants a reviewer on their side — one who checks whether they are about to hurt themselves (delete the wrong thing, leak a secret, expose something irreversibly), whether they are underselling a real win, and how the work will land with whoever decides their outcome. Also use when asked to "watch my back", "am I about to shoot myself in the foot", "am I underselling this", "protect me from myself", "defuse this before I blow up", or to run a "paladin" review. Runs INDEPENDENTLY of its sibling nemesis-review — reach for the paladin when the risk is to YOUR outcome (a footgun, an under-sell, an irreversible or public step); there is no mandatory pairing, pick by the risk.
- ▌
- ▌ Deletion Tripwire · srfinch17 bundleUse when a TRIPWIRE block message appears after a destructive command, when wiring the deletion guard on a new machine, or before any bulk deletion or cleanup — "remove junk", "clean up", "delete old files", "wipe this folder", clearing an old installation, or any recursive/force delete — so the enumerate-confirm-ledger protocol runs before anything leaves existence.
- ▌
- ▌ Training The Reader · srfinch17 bundleUse when a teaching or study artifact must get plainer across repeated cold reads; when the same reader stalls keep recurring page after page (a term used before it is defined, a number with no origin, a 40-word sentence, a picture that decorates instead of carrying the mechanism); when a rewrite must keep code, figures and tables frozen while only the prose changes; or when cold reads and Feynman-style rewrites are being delegated to cheaper models and each read should start from what the last one found instead of re-deriving it.
- ▌ Look Driven Iteration · srfinch17Use when building anything whose success is judged by eye rather than by tests - generative or procedural visual output, UI look and feel, layout, plots, 3D geometry, game feel, motion. Sets up a cheap render / one-click grounded feedback / measure-then-fix loop so the human's taste reaches you at low cost and you never burn expensive verification on a look they have not approved. Also use when tempted to review, harden or test-drive visual work before the person has seen it, when a look complaint has survived two or more fix attempts, or when asked whether agents can judge aesthetics.
- ▌ Guarding Silent Failures · srfinch17Use when an operation can fail while still returning valid-looking output - CAD/geometry booleans, image and mesh pipelines, renderers, cleanup and migration scripts, bulk edits, data transforms - or when writing the check that is supposed to catch such a failure. Also use when a test, grep, audit or probe comes back clean and that clean result is about to be trusted, when a fix "worked" but nothing visibly changed, and whenever choosing between fixing a magic number and installing an assertion.
- ▌ Managing Assumption Debt · srfinch17 bundleUse when collaborating with a human over many sessions and unstated assumptions could silently compound into expensive misunderstandings - when a rule or preference is stated in absolute terms (never/always/all/every), when about to deviate from a stated preference, when forming a belief about what the collaborator knows or has built, before producing anything irreversible or externally-facing (resumes, public sites, published repos, sends), or when checking whether you and they still agree on where the project actually stands.
- ▌
- ▌ Capturing Judgment As Data · srfinch17Use when correctness is a human's opinion rather than something a test can decide - visual or generative output, tone, taste, UX feel, "does this read as natural" - and their verdicts should survive the session. Also use when feedback keeps arriving as comments that cannot be reproduced, when a fix was declared verified against an instrument the human cannot see, when planning to carry learnings into a successor project, or when asked to build an agent that judges or screens output.
- ▌ Stretching Frontier Tokens · srfinch17Use when the strongest available model is usage-capped or rate-limited and its quota is the bottleneck on real work - "I only get a day or two before my limit resets", burning frontier tokens on mechanical work, transcripts dominated by screenshots or tool round-trips, or planning how to split a working loop across model tiers. Also use before spawning subagents to save quota, to decide what must NOT be delegated.
- ▌ Iterative Lesson Refinement · srfinch17 bundleUse when a teaching or study artifact (HTML page, study pack, explainer, interview-prep doc) must hold up under live questioning, when a lesson "reads fine" but the learner keeps failing follow-up questions, when asked to "battle-test / tighten / refine" a lesson, or when a learner needs to genuinely own a topic fast instead of reciting it.
- ▌ Rehearsing Against A Counterparty · srfinch17 bundleUse when a high-stakes written exchange is imminent and you get roughly one shot at the opening move - a negotiation, an offer response, a counter, an escalation, a dispute letter, a pitch - and the other side has a documented behavioural record. Also use when asked to "game this out", "war-game it", "run it past a simulated X", "what will they do when I send this", or when a review keeps passing but the plan has never been tested against a motivated opponent.