EAP

Apply EAP self-update from the CLI: fetch GitHub 0p9b/EAP, refresh the checkout (ff-only branch or latest release tag), record install-state, and re-run the installer. Use when the user says "eap update", "update EAP", "pull latest EAP", or runs /eap-update. Distinct from plan-only eap_upgrade.

Plan-only EAP-Runtime self-update via eap_upgrade: reports current version, resolves a pinned release tag, migrates/checks the store, re-runs doctor, and returns an apply plan — nothing is fetched or executed automatically. Use when the user says "eap upgrade", "upgrade runtime", "check for EAP release", or runs /eap-upgrade. To *apply*, tell them to run CLI `eap update` (or /eap-update).

Quick-reference card for EAP-Lean: the levels (lite/full/ultra/off), the six skills (mode + tooling), the switch syntax, defaults env/config, and how to deactivate. One-shot display, not a persistent mode. Use when the user says "lean help", "what lean commands", "how do I use eap-lean", or runs /eap-lean-help.

Forces the laziest solution that actually works — simplest, shortest, most minimal code that is still correct and safe. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra, off. Use on ANY coding task: writing, adding, refactoring, fixing, reviewing, or designing code, and choosing libraries or dependencies. Also use whenever the user says "eap lean", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", or complains about over-engineering, bloat, boilerplate, or unnecessary dependencies. Do NOT use for non-coding requests (general knowledge, prose, translation, summaries, recipes).

Health-check the EAP-Runtime layer. Calls the eap_doctor MCP tool and presents its report: node version, node:sqlite + FTS5/trigram availability, per-language runtime availability for the polyglot executor, and store integrity. Flags failures with the concrete fix (e.g. Node >= 22 for node:sqlite — there is deliberately no fallback). Use when the user says "eap doctor", "is EAP healthy", "check the runtime store", "why is eap_execute failing", or runs /eap-doctor.

Index a file, directory, or inline blob into the EAP-Runtime full-text store via eap_index. Returns pointer descriptor(s) for later eap_search. Use when the user says "index this", "eap index", "add to the offload store", or runs /eap-index. Prefer over dumping large trees into the chat.

Routing skill for EAP-Runtime: when to eap_execute vs eap_search vs eap_fetch (or eap_fetch_and_index / eap_index). Use when the agent is about to dump a large file, scrape a URL, or mine offloaded output; when the user says "think in code", "offload this", "use the runtime", or runs /eap-runtime. Prefer this over raw Read/Bash/WebFetch for oversized or network work.

Retrieve exact chunks from the EAP-Runtime offload store. Calls the eap_search MCP tool (lossless, RRF-fused keyword + substring retrieval) and presents the matching chunks with their document id, chunk index, and locator snippet — never a summary in place of the chunk body. Use when the user says "eap search <query>", "search the offloaded output", "find that in the pointer", references a pointer id like eap_ab12…, or runs /eap-search.

Report what EAP-Runtime has kept out of the context window this session, in MEASURED terms only. Calls the eap_stats MCP tool and presents its numbers: documents indexed, chunks, bytes kept out (a real sum of indexed bytes), and the labelled ~bytes/4 token estimate. Never invents percentages or dollar figures. Use when the user says "eap stats", "how much has EAP saved", "bytes kept out", "runtime stats", or runs /eap-stats.

Ultra-compressed commit message generator. Conventional Commits. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/commit", or /eap-signal-commit.

Compress natural language memory files (CLAUDE.md, todos, preferences) into EAP-Signal prose to save input tokens. Preserves code, URLs, paths, structure. Backup under XDG data dir (or %LOCALAPPDATA%). Trigger: /eap-signal-compress FILEPATH or "compress memory file".

Quick-reference card for EAP-Signal modes, skills, and commands. One-shot display, not a persistent mode. Trigger: /eap-signal-help, "signal help", "what signal commands", "how do I use EAP-Signal".

Ultra-compressed code review comments. One line per finding: location, problem, fix. Use when user says "review this PR", "code review", "review the diff", "/review", or /eap-signal-review.

Verdict-first communication mode for EAP-Signal. Cuts filler while preserving full technical accuracy. Levels: lite, full (default), ultra, wenyan-lite, wenyan-full (alias: wenyan), wenyan-ultra, off. Use when user says "signal mode", "tldr mode", "talk TLDR", "verdict first", "no filler", or invokes /eap signal / /eap-signal.

Show measured session prose/token stats for EAP-Signal. Reads the Claude Code session transcript when available — no invented savings percentages. Triggers on /eap-signal-stats. Prefer the hook-injected report when wired.

Decision guide for delegating to EAP-Signal-compressed subagents. When to spawn eapcrew-investigator (locate), eapcrew-builder (1-2 file edit), or eapcrew-reviewer (diff review) instead of vanilla Explore. Compressed tool results keep more of main-context budget. Trigger: "delegate to subagent", "use eapcrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".

One-shot impact scoreboard for EAP-Lean — measured values only. Counts the `eap-lean:` markers actually in this repo and aggregates any "net: -N lines" figures the review/audit skills produced this session. NEVER prints a benchmark percentage, an upstream headline number, or a per-repo cost claim: the unbuilt version was never written, so there is no baseline to subtract from. One-shot display, not a persistent mode. Use when the user says "lean gain", "what does lean save", "show lean impact", "lean scoreboard", or runs /eap-lean-gain.

Maintenance: clear the EAP-Runtime offload store, or drop a single indexed document. Calls the eap_purge MCP tool — whole-store purge only after the user confirms, single-document purge when they name a pointer id. Reports exactly how many documents were removed. Use when the user says "eap purge", "clear the eap store", "drop that pointer", "reset the runtime index", or runs /eap-purge.

Whole-repo audit for over-engineering — the eap-lean-review lens applied to the entire tree instead of one diff. Produces a ranked delete-list (biggest cut first), each line tagged delete/stdlib/native/yagni/shrink with what to cut and its replacement, ending in a "net: -N lines, -M deps possible" estimate. Complexity only; correctness, security, and performance are out of scope. One-shot report — applies nothing. Use when the user says "audit this codebase", "audit for over-engineering", "what can I delete from this repo", "find the bloat", "lean audit", or runs /eap-lean-audit.

Harvest every `eap-lean:` comment in the codebase into a debt ledger, so the deliberate shortcuts EAP-Lean leaves behind stay tracked instead of rotting into "later means never". One row per marker: what was simplified, the ceiling it named, and the trigger to revisit. Flags any marker with no upgrade path as a silent-rot risk. One-shot report — changes nothing unless asked to persist it. Use when the user says "lean debt", "what did lean defer", "list the shortcuts", "lean ledger", "what did we mark to do later", or runs /eap-lean-debt.

Over-engineering review of a diff — nothing else. Emits a delete-list: one line per finding, each tagged delete/stdlib/native/yagni/shrink, giving the location, what to cut, and what replaces it. Ends with a single "net: -N lines possible" estimate. Complexity only — correctness bugs, security holes, and performance are explicitly OUT of scope; route those to a normal review. Lists findings, applies nothing. Use on a code change when the user says "review for over-engineering", "what can we delete", "is this over-built", "lean review", or runs /eap-lean-review.