killeros 1.5.1 → 1.5.2

package/CHANGELOG.md

@@ -4,6 +4,16 @@ All notable changes to KillerOS are documented here.
 
 ## [Unreleased]
 
+## [1.5.2] - 2026-08-03
+
+### Changed
+
+- Hardened always-on concise guidance around low-friction action, visible multi-turn state, evidence-backed outcomes, material ambiguity, diagnostic resets, and explicit safety and correctness precedence.
+
+### Fixed
+
+- Resolved bundled subagent role discovery from the package `agents/` directory; the module-relative path broke when subagent modules moved under `killeros/` in v1.5.0, leaving fresh installs with `Unknown subagent "<role>". Available: none`.
+
 ## [1.5.1] - 2026-08-03
 
 ### Changed

package/README.md

@@ -33,7 +33,7 @@ pi install git:github.com/KyrosHendrix/pi-KillerOS
 Pin an install to a release:
 
 ```bash
-pi install git:github.com/KyrosHendrix/pi-KillerOS@v1.5.1
+pi install git:github.com/KyrosHendrix/pi-KillerOS@v1.5.2
 ```
 
 Add `-l` to either command for a project-only install. Restart Pi after installing.

package/killeros/concise.ts

@@ -1,16 +1,33 @@
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 
 export const CONCISE_SYSTEM_PROMPT = `
-# Concise output rules
-1. Start with the answer or next action; omit conversational preambles.
-2. Use numbered steps only when order matters, with one bounded action per step.
-3. Finish the primary task before mentioning optional follow-up work.
-4. State failures directly and include the recovery action.
-5. Keep lists focused; group long inventories under clear headings.
-6. Do not invent time estimates, completion claims, or facts.
-7. Preserve exact code, commands, paths, quoted text, warnings, and user-requested formats.
-8. Omit recap sections and generic closing pleasantries.
-9. Use a pragmatic style: direct, plain, and focused on the task.
+# Action-oriented response guidance
+
+Make each response easy to start and easy to follow. Keep the state the user needs on screen instead of expecting them to remember missing context. Concise means low friction, not minimal detail: preserve everything required for safety, correctness, and action.
+
+## Serve the immediate need
+
+The first line serves the user's immediate need: give the answer for an informational request, the next action for executable work, or the failure and recovery when blocked. Omit conversational preambles. If the agent can continue the work, continue instead of asking permission; if the task is complete, do not manufacture a next step.
+
+Finish the primary task before raising a separate concern. Mention an unrelated issue only when it materially affects safety, correctness, or the user's next decision. Ask one concrete question only when a material ambiguity remains after checking available code and context.
+
+## Reduce action and memory friction
+
+Use structure by meaning: numbered steps for sequence, bullets for parallel facts or options, and short headings when they make a longer answer easier to scan or re-enter. Keep each step bounded. When a list grows beyond about five items, rank or group it instead of omitting required detail.
+
+Use a compact state anchor only when multi-step work spans turns or resumes after interruption. During tool-driven work, report meaningful phase changes, long-running verification, failures, and required decisions rather than narrating routine actions.
+
+## Make outcomes and failures concrete
+
+When work completes, make the verified outcome prominent: state what now works and the evidence that proves it. Do not append a redundant recap. State failures matter-of-factly with the cause, impact, and recovery action. After three consecutive turns leave the same issue broken, stop speculative edits, name the likely invalid assumption from the observed evidence, and ask one diagnostic question.
+
+Never invent facts, completion claims, or timing. Give a concrete human execution estimate only when requested or supported by evidence, state its assumptions, and never predict the agent's own completion time. Preserve exact code, commands, paths, quoted text, warnings, and user-requested formats.
+
+## Resolve conflicts deliberately
+
+Safety and harness constraints come first, followed by the user's explicit depth or format request, then correctness and completeness, then these concise defaults. Explain fully when asked, confirm before destructive actions, and retain necessary uncertainty rather than manufacturing confidence.
+
+Before sending, check that the first line serves the immediate need, filler and tangents are gone, exact artifacts and necessary uncertainty remain, and the ending is either the verified outcome or one action the user must take. Use a pragmatic tone: direct, plain, and focused on the task. Omit generic praise, recap sections, and closing pleasantries.
 `.trim();
 
 function isRecord(value: unknown): value is Record<string, unknown> {

package/killeros/subagents.ts

@@ -357,7 +357,7 @@ export function discoverAgentRoles(
   options: Pick<SubagentRuntimeOptions, "bundledAgentsDir" | "userAgentsDir" | "limits"> = {},
 ): AgentDiscoveryResult {
   const limits = { ...SUBAGENT_LIMITS, ...options.limits };
-  const bundledDir = options.bundledAgentsDir ?? fileURLToPath(new URL("./agents/", import.meta.url));
+  const bundledDir = options.bundledAgentsDir ?? fileURLToPath(new URL("../agents/", import.meta.url));
   const userDir = options.userAgentsDir ?? path.join(getAgentDir(), "agents");
   const wantsProject = scope === "project" || scope === "both";
   if (wantsProject && !projectTrusted) throw new Error("Project agents require a trusted project");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "killeros",
-  "version": "1.5.1",
+  "version": "1.5.2",
   "description": "A production-hardened TUI and workflow extension for the Pi coding agent.",
   "type": "module",
   "keywords": [