@prestyj/cli 5.28.1 → 5.29.0

package/dist/core/compaction/plan-step-policy.js

@@ -0,0 +1,57 @@
+/**
+ * Plan-step compaction policy (port of NVlabs/SoL-Pi commit 4101906, as
+ * re-implemented in bench/baseline/32-plan-step-compaction.mjs arm B).
+ *
+ * Checked at turn end after a plan step is newly completed. Compacting turns a
+ * big cached prefix into a small summary: every later request re-reads less
+ * (saving), but the new prefix must be re-written to cache once at r× the read
+ * price. Compact iff that write cost is repaid within the expected number of
+ * remaining requests (horizon). Pure — the session owns the bookkeeping.
+ */
+/** SoL-Pi default cache write/read price ratio. GPT-6.1 Sol's real ratio is 25. */
+export const DEFAULT_CACHE_WRITE_READ_RATIO = 12.5;
+/** Recent tail kept verbatim by compaction (SoL-Pi keep budget). */
+export const PLAN_STEP_KEEP_TOKENS = 20_000;
+/** Approximate size of the summary that replaces the archive. */
+export const PLAN_STEP_SUMMARY_TOKENS = 1_000;
+/** Within this many tokens of the window, leave it to the size trigger. */
+export const PLAN_STEP_WINDOW_GUARD = 16_384;
+export function decidePlanStepCompaction(input) {
+    const ctx = input.contextTokens;
+    if (input.stepsRemaining <= 0)
+        return { compact: false, reason: "plan-finished" };
+    if (ctx >= input.contextWindow - PLAN_STEP_WINDOW_GUARD) {
+        return { compact: false, reason: "near-window: defer to size trigger" };
+    }
+    if (input.priorCompactions > 0 && (input.requestsSinceLastCompaction ?? 1) < 1) {
+        return { compact: false, reason: "back-to-back" };
+    }
+    const r = Math.max(1, input.cacheWriteReadRatio);
+    const archive = Math.max(0, ctx - input.keptTailTokens);
+    const saving = archive - PLAN_STEP_SUMMARY_TOKENS;
+    if (saving <= 0)
+        return { compact: false, reason: "no-saving" };
+    const post = ctx - saving;
+    const breakeven = (post * (r - 1)) / saving;
+    const done = input.stepsCompleted;
+    const meanReq = done > 0 ? input.requestsInCompletedSteps / done : 0;
+    let horizon = 1 + Math.floor(meanReq * input.stepsRemaining);
+    const growth = input.requests > 0 ? input.grownTokens / input.requests : 0;
+    if (growth > 0) {
+        horizon = Math.min(horizon, Math.floor((input.contextWindow - ctx) / growth));
+    }
+    if (input.priorCompactions === 0) {
+        horizon *= 2;
+        return {
+            compact: breakeven <= horizon,
+            reason: `first be=${breakeven.toFixed(1)} h=${horizon}`,
+        };
+    }
+    // Later compactions: stricter margin, and earlier write cost must be repaid.
+    const debtCovered = input.writeCostBalance <= 0;
+    return {
+        compact: breakeven * 1.5 <= horizon && debtCovered,
+        reason: `be=${breakeven.toFixed(1)} h=${horizon} debt=${Math.round(input.writeCostBalance)}`,
+    };
+}
+//# sourceMappingURL=plan-step-policy.js.map

package/dist/core/compaction/plan-step-policy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"plan-step-policy.js","sourceRoot":"","sources":["../../../src/core/compaction/plan-step-policy.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,mFAAmF;AACnF,MAAM,CAAC,MAAM,8BAA8B,GAAG,IAAI,CAAC;AACnD,oEAAoE;AACpE,MAAM,CAAC,MAAM,qBAAqB,GAAG,MAAM,CAAC;AAC5C,iEAAiE;AACjE,MAAM,CAAC,MAAM,wBAAwB,GAAG,KAAK,CAAC;AAC9C,2EAA2E;AAC3E,MAAM,CAAC,MAAM,sBAAsB,GAAG,MAAM,CAAC;AA+B7C,MAAM,UAAU,wBAAwB,CACtC,KAA8B;IAE9B,MAAM,GAAG,GAAG,KAAK,CAAC,aAAa,CAAC;IAChC,IAAI,KAAK,CAAC,cAAc,IAAI,CAAC;QAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,eAAe,EAAE,CAAC;IAClF,IAAI,GAAG,IAAI,KAAK,CAAC,aAAa,GAAG,sBAAsB,EAAE,CAAC;QACxD,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,oCAAoC,EAAE,CAAC;IAC1E,CAAC;IACD,IAAI,KAAK,CAAC,gBAAgB,GAAG,CAAC,IAAI,CAAC,KAAK,CAAC,2BAA2B,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QAC/E,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,cAAc,EAAE,CAAC;IACpD,CAAC;IACD,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,mBAAmB,CAAC,CAAC;IACjD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,KAAK,CAAC,cAAc,CAAC,CAAC;IACxD,MAAM,MAAM,GAAG,OAAO,GAAG,wBAAwB,CAAC;IAClD,IAAI,MAAM,IAAI,CAAC;QAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;IAChE,MAAM,IAAI,GAAG,GAAG,GAAG,MAAM,CAAC;IAC1B,MAAM,SAAS,GAAG,CAAC,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC;IAC5C,MAAM,IAAI,GAAG,KAAK,CAAC,cAAc,CAAC;IAClC,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,wBAAwB,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACrE,IAAI,OAAO,GAAG,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,GAAG,KAAK,CAAC,cAAc,CAAC,CAAC;IAC7D,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,WAAW,GAAG,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3E,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,aAAa,GAAG,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC;IAChF,CAAC;IACD,IAAI,KAAK,CAAC,gBAAgB,KAAK,CAAC,EAAE,CAAC;QACjC,OAAO,IAAI,CAAC,CAAC;QACb,OAAO;YACL,OAAO,EAAE,SAAS,IAAI,OAAO;YAC7B,MAAM,EAAE,YAAY,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,OAAO,EAAE;SACxD,CAAC;IACJ,CAAC;IACD,6EAA6E;IAC7E,MAAM,WAAW,GAAG,KAAK,CAAC,gBAAgB,IAAI,CAAC,CAAC;IAChD,OAAO;QACL,OAAO,EAAE,SAAS,GAAG,GAAG,IAAI,OAAO,IAAI,WAAW;QAClD,MAAM,EAAE,MAAM,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,OAAO,SAAS,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,gBAAgB,CAAC,EAAE;KAC7F,CAAC;AACJ,CAAC"}

package/dist/core/destructive-git-guard.d.ts

@@ -0,0 +1,90 @@
+/**
+ * Destructive-git guard: refuses model-issued shell commands that would throw
+ * away the user's uncommitted work.
+ *
+ * Checkpoints only capture edits made through write/edit, so one
+ * `git reset --hard` from the shell would otherwise lose work with no undo.
+ *
+ * Two phases:
+ * 1. {@link findDestructiveGitCommands} — pure parse. A small quote-aware shell
+ *    scanner splits the command into simple commands (through `&&`/`;`/`||`/
+ *    pipes, subshells, `$(…)`/backticks, `if`/`for`/`while` bodies), unwraps
+ *    `bash -c`/`sh -c`/`eval` scripts and exec wrappers (`sudo`, `env`,
+ *    `timeout`, `nohup`, `xargs`, …), follows `cd` and `git -C`, and classifies
+ *    each git invocation.
+ * 2. {@link checkDestructiveGit} — context. Discarding commands are blocked
+ *    only when `git status` in the target directory shows something they would
+ *    destroy; a clean tree passes. Plain force-push is always blocked.
+ *
+ * Fails OPEN on guard errors (git missing, timeout): the guard is a seatbelt
+ * against an agent mistake, not a sandbox, and must never wedge the bash tool.
+ */
+type DestructiveGitKind = "reset-hard" | "checkout-paths" | "restore-worktree" | "clean" | "stash-drop" | "stash-clear" | "branch-force-delete" | "force-push";
+export interface DestructiveGitMatch {
+    kind: DestructiveGitKind;
+    /** The git invocation as parsed, for the block message. */
+    display: string;
+    /** Directory git would run in (after `cd` and `git -C`). */
+    dir: string;
+    /** `--git-dir=`/`--work-tree=` globals to forward to inspection commands. */
+    gitArgs: string[];
+    /** Pathspecs limiting what is discarded; empty means the whole tree. */
+    pathspecs: string[];
+    /** checkout with one operand that may be a branch rather than a path. */
+    maybeRef?: string;
+    /** clean: also/only removes ignored files (`-x` / `-X`). */
+    ignored?: "also" | "only";
+    /** clean: `-d` also removes untracked directories. */
+    directories?: boolean;
+    /** branch -D targets; stash drop target. */
+    refs: string[];
+}
+/** Basename of a command word, `.exe` stripped (`/usr/bin/git`, `git.exe`). */
+export declare function commandName(word: string): string;
+/**
+ * Expand the home and working-directory forms a model writes in paths:
+ * `~`, `$HOME`/`${HOME}` (any case, which also covers PowerShell `$home`),
+ * `$env:USERPROFILE`/`$env:HOME`, cmd `%USERPROFILE%`, and `$PWD`/`${PWD}`. Other variables stay
+ * literal, so they resolve to a harmless relative path.
+ */
+export declare function expandShellPath(p: string, cwd: string): string;
+/** One simple command found in a shell string, after unwrapping. */
+export interface ShellInvocation {
+    /** {@link commandName} of the command word. */
+    name: string;
+    args: readonly string[];
+    /** Directory it would run in after any earlier `cd` in the same script. */
+    cwd: string;
+    /** Run by `xargs`, which appends operands read from stdin. */
+    viaXargs: boolean;
+}
+/**
+ * Every simple command in `command` whose name satisfies `wanted`, in source
+ * order, after splitting on `&&`/`;`/`||`/pipes, entering subshells,
+ * substitutions and `if`/`for`/`while` bodies, unwrapping `bash -c`/`eval`/
+ * `pwsh -c` scripts and exec wrappers (`sudo`, `env`, `timeout`, `xargs`, …),
+ * and following `cd`. Pure: nothing is executed.
+ */
+export declare function walkShell(command: string, cwd: string, wanted: (name: string) => boolean): ShellInvocation[];
+/**
+ * Parse `command` and return every git invocation in it that can destroy
+ * work, with the directory each would run in. Pure apart from `existsSync`
+ * for ambiguous `git checkout <name>`.
+ */
+export declare function findDestructiveGitCommands(command: string, cwd: string): DestructiveGitMatch[];
+export interface DestructiveGitCheckOptions {
+    /** The tool's working directory. */
+    cwd: string;
+    /**
+     * The shell's real current directory when it can differ from `cwd`
+     * (persistent session shell after a `cd`). Only called when the command
+     * contains a destructive git invocation.
+     */
+    resolveCwd?: () => Promise<string | null>;
+}
+/**
+ * Returns a block message for the model when `command` would destroy
+ * uncommitted work (or force-push), or null to let it run.
+ */
+export declare function checkDestructiveGit(command: string, options: DestructiveGitCheckOptions): Promise<string | null>;
+export {};