thiamine

Plugin thiamine from marketplace thiamine-stack in vihang96/thiamine-stack. Commands read from the repo's manifest — 6 days ago.

Skills (22)

• branch-to-pr

  Decides where a change lands and carries it from there to a pull request ready for approval, in one repo or several. Use before the first edit of any change that will be committed, when work touches more than one service, when creating or cleaning up worktrees, when pausing or resuming a session, wh

• capture-preferences

  Captures how a person likes to work as a portable <handle>-mode skill that agents load and follow, covering response style, autonomy, verification posture, delegation, and the corrections they keep making. Use for automate me, capture or update my working style, make a mode skill for me, when the sa

• consistency

  Finds the answer a codebase already has for a concern before adding another one, and records the decision in an ADR when there is none. Use when introducing a pattern, choosing a library, a database, or a store, choosing a dev or test tool such as a tunnel, a browser driver, or a local stack, portin

• continual-learning

  Runs the continual-learning loop by delegating transcript mining to the memory-updater agent, keeping the mined transcripts out of the main context. Use when asked to mine prior chats, refresh project memory, or remember what was learned in earlier sessions.

• experimentation

  Runs a sustained improvement loop against a measured objective: states the goal, isolates the one metric being moved from the ones that must only be held, generates hypotheses from clustered failures, and prunes what stops paying. Use for hill climbing, tuning prompts for accuracy or overlap scores,

• fan-out-work

  Runs several subagents in parallel from one session and converges what they return into one coherent change. Use when work splits into units that can run at once, when spawning subagents to investigate or write code in parallel, when a change spans several repos that can be parallelised and worked s

• handoff

  Keeps a durable record of work in progress so a new session can continue it, reconstructs that record from transcripts and live state when none was kept, and prunes it once the work it describes has landed. Use when starting work that will span sessions, when recording a decision or a deviation mid-

• interrogate

  Reviews an artifact somebody else produced, a human or an agent, and returns feedback precise enough to act on: a plan, a design, a diff, a pull request, or an agent's report. Decides what earns a finding, ranks what is left, stops before the nits, and says plainly when there is nothing to report. U

• maintain-skills

  Runs a periodic pass over a repo's agent-facing context (rules, skills, AGENTS.md, CLAUDE.md) to find guidance that no longer matches the code, and routes each finding to a fix or a check. Use when asked to audit, refresh, or spring-clean the skills, when the feature map has outgrown what agents are

• post-implementation

  Closes out a change after the code is written and before it lands: explaining it so the user can defend it in review, confirming each decision was the one they wanted, and getting buy-in from teammates with a summary, screenshots, or a demo. Use when a change is finished, when a turn both lands work

• pr-review-sweep

  Sweeps recent open GitHub pull requests for missing independent reviews, verifies actionable findings, and follows reviewed PRs through fixes and merge confidence. Use for a PR review sweep, a daily review watcher, watcher cleanup and retention, or re-reviewing fixes from an earlier sweep.

• pre-implementation

  Prepares work before any code is written: sorting unknowns into what to observe and what to ask, understanding the code that exists, sketching a prototype to settle a design or a domain question, choosing the metric a performance push is measured by, finding the blast radius of what breaks if you ch

• reflect

  Mines a finished session, or several, for what was learned the hard way and encapsulates it as a skill, agent definition, rule, or check so the next agent does not rediscover it. Use when asked to reflect on a session, to capture what a long build taught, to make a hard-won approach repeatable, or t

• signal-to-task

  Carries a signal from a running system to a disposition: an alert, a crash group, a trace or a profile, a slow endpoint, a failed run, a complaint in a channel, a ticket nobody has read. Reduces it to one claim with evidence, diagnoses it from a capture or a live process, decides whether it earns wo

• technical-writing

  Layered technical-writing standard: Diátaxis structure, Google developer style sentences, STE instruction rules, Global English syntax. Use when writing or reviewing docs, RFCs, readmes, decision records, design docs, test plans, PR descriptions, or commit messages, including docs written along the

• thiamine-author

  Authors or revises a thiamine skill, agent, command, or rule from the repo templates, picking the right artifact type and shape and validating it before finishing. Use when adding to or changing the thiamine stack, when asked to add a skill, rule, or command for engineering standards, for should thi

• thiamine-install

  Installs, updates, or removes the thiamine engineering standards across coding harnesses (Claude Code, Codex, Cursor) by symlinking skills and wiring the rules file into each harness's global instructions. Use when setting up thiamine on a new machine, adding a harness, wiring thiamine into a repo,

• unslop-prose

  Cuts AI tells from prose: puffery, AI vocabulary, filler, hedging, em dashes, inline-header lists, and sycophantic chatter. Use when writing or editing docs, PR descriptions, commit messages, code comments, or chat replies, and when text reads as machine-generated.

• unslop-python

  Rejects low-evidence Python: Any in signatures, bare except, mutable default arguments, commented-out code, print in shipped code, and tests that assert on mocks. Assumes uv and ruff. Use when reading or editing any .py file, when reviewing a Python diff, or when setting up a Python project or addin

• unslop-rust

  Rejects low-evidence Rust: unwrap, swallowed errors, undocumented unsafe, tautological tests, meta design docs, and C-shaped code ported into Rust. Use when reading or editing any .rs file, when reviewing a Rust diff, when fighting the borrow checker into compiling, or when a Cargo crate is being se

• unslop-typescript

  Rejects low-evidence TypeScript: as casts, unknown in contracts, optional-field bags, broad dictionary types, module mocks, and runtime typeof narrowing. Use when reading or editing any .ts or .tsx file, when reviewing a TypeScript diff, or when types are being loosened to make an error go away.

• working-alongside

  Decides whether work about to start will clash or interfere with work another session or another person already has in flight, then starts it, starts it expecting merge conflicts, or arms it to begin when the other work lands. Use when another session is already running on the repo, when opening a s

Subagents (5)

• adversarial-reviewer

  Reviews an artifact it did not write, from one assigned angle, and returns only defects it can trace to a concrete failure. Use to spawn one reviewer per angle over a diff, a plan, or a report, when review needs a context that did not produce the work.

• code-simplifier

  Use this agent when you have functional code that needs refactoring to improve readability, reduce complexity, or eliminate redundancy. Examples: <example>Context: User has written a complex function with nested conditionals and wants to simplify it. user: "Here's my authentication function with mul

• coherence-reviewer

  Reads a summary of several parallel lanes' work and reports only the defects that exist between them: the same thing added twice, one question answered two ways, an invented stub standing in for another lane's contract, and naming that drifted. Use after parallel subagents or parallel sessions finis

• curator

  Curates the failing cases of an improvement run into the few groups worth acting on, grouped by what went wrong, each with a count, the share of the gap it caps, and a mechanism where the evidence shows one. Use during an improvement run when there are more failures than fit in one context, and befo

• memory-updater

  Mines new agent transcript entries for durable user preferences and project facts, then writes them into the harness's own memory store. Use when asked to mine prior chats, refresh project memory, or run the continual-learning loop.

Hooks (6)

• PreToolUse

  node "${CLAUDE_PLUGIN_ROOT}/hooks/pre-edit-branch-guard.mjs"

• PostToolUse

  node "${CLAUDE_PLUGIN_ROOT}/hooks/post-edit-validate.mjs"

• PostToolUse

  node "${CLAUDE_PLUGIN_ROOT}/hooks/pr-body-budget.mjs"

• Stop

  node "${CLAUDE_PLUGIN_ROOT}/hooks/stop-count-turns.mjs"

• SessionStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/session-start-lanes.mjs"

• SessionStart

  node "${CLAUDE_PLUGIN_ROOT}/hooks/session-start-suggest.mjs"

Similar plugins