@olegkoval/agent-skills 1.42.0 → 1.42.2

package/adapters/claude/olko-release/skills/changelog-generator/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: changelog-generator
-description: Automatically creates user-facing changelogs from git commits by analyzing commit history, categorizing changes, and transforming technical commits into clear, customer-friendly release notes. Turns hours of manual changelog writing into minutes of automated generation.
+description: >
+  Turn git commit history into a user-facing changelog: read the commits since the last release,
+  categorise them, and rewrite technical messages as release notes a customer can actually read.
+  Use when the user asks for "a changelog", "release notes", "what changed since the last release",
+  "write the CHANGELOG entry", or is preparing a release and needs the notes written. For setting
+  up the release pipeline itself use semantic-release-beta.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires git history access in the target repository.

package/adapters/claude/olko-release/skills/semantic-release-beta/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: semantic-release-beta
-description: Set up or update a Node.js release pipeline that uses semantic-release-npm-github-publish with a stable main channel and a prerelease beta channel on a beta branch, including GitHub Actions validation and npm publishing behavior.
+description: >
+  Set up or repair a Node.js release pipeline built on semantic-release-npm-github-publish, with a
+  stable channel on main and a prerelease channel on a beta branch, including the GitHub Actions
+  validation and npm publishing behaviour. Use when the user wants "semantic release set up", "a
+  beta release channel", "prerelease publishing", "automated npm publishing", or when an npm
+  package's releases are still manual and should become commit-driven. For writing the notes of a
+  single release use changelog-generator.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires a Node.js package repository that uses git and GitHub Actions.

package/adapters/claude/olko-skill-meta/skills/ai-tools-setup/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-tools-setup
-description: 'Set up, repair, and report on the RTK+ICM+Vox AI development toolkit. Installs missing tools, fixes broken hooks and MCP config, and shows an effectiveness digest (token savings, memory accumulation, voice health) on demand or as a scheduled weekly report.'
+description: >
+  Set up, repair, and report on the RTK+ICM+Vox AI development toolkit: installs missing tools,
+  fixes broken hooks and MCP config, and shows an effectiveness digest covering token savings,
+  memory accumulation and voice health. Use when the user says "set up my AI tools", "my hooks are
+  broken", "MCP is not loading", "check my toolkit", "is ICM working", "how much are these tools
+  saving me", or when the scheduled weekly toolkit report fires.
 license: MIT
 allowed-tools: Bash, Read, Edit, Write
 compatibility: Claude Code on macOS (Homebrew required)

package/adapters/claude/olko-skill-meta/skills/promptctl/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: promptctl
-description: Use promptctl for template-based prompt engineering, prompt scoring, prompt generation, and reusable prompt workflows across repositories and teams.
+description: >
+  Manage reusable prompts with the promptctl CLI: template-based prompt engineering, prompt
+  generation, and scoring a prompt before it ships. Use when the user wants to "write a reusable
+  prompt", "score this prompt", "turn this into a prompt template", "share prompts across repos",
+  or names promptctl directly. For bringing an agent skill into this catalog use
+  add-to-my-skills.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires promptctl to be installed when the workflow depends on the CLI.

package/adapters/claude/olko-web-ops/skills/docs-index-keeper/SKILL.md

@@ -1,6 +1,10 @@
 ---
 name: docs-index-keeper
-description: Keep a repository's Markdown docs index in sync by installing and running docs-index-keeper in pre-commit hooks, CI, or one-off maintenance flows.
+description: >
+  Keep a repository's Markdown docs index in sync with the files that actually exist, by installing
+  and running docs-index-keeper in a pre-commit hook, in CI, or as a one-off sweep. Use when the
+  user says "the docs index is stale", "update the docs index", "add a docs index check to CI",
+  "our doc links are out of date", or when new docs were added without being linked anywhere.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit, Grep, Glob
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires shell access and Node.js 18+ when installing from npm.

package/adapters/claude/olko-web-ops/skills/website-analytics-bootstrap/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: website-analytics-bootstrap
-description: Set up a website's first SEO analytics loop end-to-end: persistent SerpBear rank tracking, Google Search Console, seeded keywords, read-only technical audits, and Telegram alerts on a local host or NAS.
+description: >
+  Stand up a website's first SEO analytics loop end to end: persistent SerpBear rank tracking,
+  Google Search Console, seeded keywords, read-only technical audits and Telegram alerts on a local
+  host or NAS. Use when the user says "set up rank tracking", "this site has no analytics", "track
+  my keywords", "bootstrap SEO monitoring", or wants to know how a site performs in search when no
+  tracking exists yet. For diagnosing an existing indexing problem use
+  search-console-indexing-audit.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires network access, a reachable deployment host, and owner-provided Search Console and alert credentials.

package/adapters/cursor/olko-creative/skills/vinted-listing/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: vinted-listing
-description: Create, edit, and safely publish Vinted listings through the signed-in browser, using verified item details and the owner’s original photos.
+description: >
+  Create, edit, and safely publish a Vinted listing through the signed-in browser, using verified
+  item details and the owner's original photos. Suggests title, description, category, brand,
+  condition, colour and parcel size from the photos, saves a draft first, and requires explicit
+  confirmation before publishing. Use when the user wants to list, sell, relist or price an item on
+  Vinted, or says "put this on Vinted", "create a Vinted listing", "sell this on Vinted", "edit my
+  Vinted draft", or "publish that listing". Browser workflow, not an API integration.
 ---
 <!-- Generated by scripts/build-adapters.sh. Do not edit directly. -->
 

package/adapters/cursor/olko-git-tools/skills/gh-cli/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: gh-cli
-description: GitHub CLI (gh) comprehensive reference for repositories, issues, pull requests, Actions, projects, releases, gists, codespaces, organizations, extensions, and all GitHub operations from the command line.
+description: >
+  Reach for the GitHub CLI (gh) to do something on GitHub from the command line: repositories,
+  issues, pull requests, Actions runs and logs, projects, releases, gists, codespaces, org and
+  extension management. Use when a task needs a gh command and the exact syntax, flag or API path
+  is not obvious, or the user says "use gh", "what is the gh command for", "list my PRs from the
+  CLI", or "query the GitHub API". For driving a PR through review bots and CI use the
+  olko-github-pr skills instead; this one is the command reference.
 license: MIT
 allowed-tools: Bash
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires the GitHub CLI and valid GitHub authentication.

package/adapters/cursor/olko-product/skills/viral-launch/SKILL.md

@@ -1,7 +1,11 @@
 ---
-
 name: viral-launch
-description: Set up a project repository and launch plan for shareable marketing, public launch readiness, and growth loops without pretending virality is guaranteed.
+description: >
+  Prepare a project for a public launch: shareable positioning, a launch-readiness pass over the
+  repository, and the growth loops worth building, without pretending virality is guaranteed. Use
+  when the user says "help me launch this", "prep for Product Hunt", "write the launch post", "make
+  this shareable", "launch plan", or is about to make a repo or product public. For building the
+  thing itself use mvp-oneshot or product-builder.
 license: MIT
 allowed-tools: Bash, Read, Write, WebSearch, WebFetch
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Useful for product repos, open-source packages, SaaS launches, waitlists, and creator tools.

package/adapters/cursor/olko-reflection/skills/agent-ops-retro/scripts/mine-transcripts.mjs

@@ -27,7 +27,7 @@ const OUT = arg('out', './mine.json');
 const SINCE = Date.parse(`${arg('since', '')}T00:00:00Z`);
 const UNTIL = Date.parse(`${arg('until', '')}T00:00:00Z`);
 
-if (!Number.isFinite(SINCE) || !Number.isFinite(UNTIL) || UNTIL <= SINCE) {
+if (!Number.isFinite(SINCE) || !Number.isFinite(UNTIL)) {
   console.error('usage: mine-transcripts.mjs --since YYYY-MM-DD --until YYYY-MM-DD [--out path] [--root dir]');
   process.exit(2);
 }
@@ -46,7 +46,23 @@ function walk(dir, out = []) {
 }
 
 // A user-role text record that is really harness output, not something a person typed.
-const HARNESS = /^(<system-reminder|<command-name|<command-message|<local-command|<bash-input|<bash-stdout|<user-prompt-submit-hook|<task-notification|Caveat: The messages below|\[Request interrupted)/;
+// A user-role text record that is really harness output, not something a person typed.
+// Each alternative below was found by a run that mis-counted. The first pass caught the
+// obvious wrappers and still let 26% through: skill body dumps, image attachments,
+// already-loaded stubs and permission-grant echoes all sit in human-turn position.
+const HARNESS = new RegExp('^(' + [
+  '<system-reminder', '<command-name', '<command-message', '<local-command',
+  '<bash-input', '<bash-stdout', '<bash-stderr', '<user-prompt-submit-hook',
+  '<task-notification', '<cross-session-message',
+  'Caveat: The messages below',
+  '\\[Request interrupted',
+  'Base directory for this skill:',      // a skill body dumped into the turn
+  'Skill /[^ ]+ is already loaded',      // re-invocation stub
+  'Permission granted for:',             // permission-grant echo
+  '\\[Image #\\d+\\]',                    // bare attachment
+  '\\[Image: source:',
+  '\\[Pasted text',
+].join('|') + ')');
 
 const textOf = (msg) => {
   const c = msg?.content;
@@ -68,7 +84,8 @@ const S = {
   limits: [],
   humanTurns: [],
   echoTurns: 0,
-  interrupts: 0,
+  interruptsAll: 0,
+  interruptsMain: 0,
   compacts: 0,
   usage: { in: 0, out: 0, cacheRead: 0, cacheWrite: 0 },
 };
@@ -134,16 +151,8 @@ for (const f of files) {
 
     if (r.type === 'assistant') {
       const model = r.message?.model;
+      if (model) bump(S.models, model);
       const u = r.message?.usage;
-      if (model) {
-        const stats = S.models.get(model) || { messages: 0, in: 0, out: 0, cacheRead: 0, cacheWrite: 0 };
-        stats.messages++;
-        stats.in += u?.input_tokens || 0;
-        stats.out += u?.output_tokens || 0;
-        stats.cacheRead += u?.cache_read_input_tokens || 0;
-        stats.cacheWrite += u?.cache_creation_input_tokens || 0;
-        S.models.set(model, stats);
-      }
       if (u) {
         S.usage.in += u.input_tokens || 0;
         S.usage.out += u.output_tokens || 0;
@@ -182,9 +191,12 @@ for (const f of files) {
     }
 
     const raw = textOf(r.message);
+    // Two scopes, two names. The first version reported the all-records count and the
+    // per-session count under one label `interrupts`, so the top-level figure (27) did not
+    // match the sum of the per-session rows (15) and neither number could be trusted.
     if (raw.includes('[Request interrupted')) {
-      S.interrupts++;
-      if (sess) sess.interrupts++;
+      S.interruptsAll++;
+      if (sess) { S.interruptsMain++; sess.interrupts++; }
     }
     if (!main || r.userType !== 'external' || !raw || sawToolResult) continue;
 
@@ -219,13 +231,14 @@ const report = {
   humanTurnsPerDay: byDay,
   turnsPerSession: { median: pct(0.5), p90: pct(0.9), max: humanCounts.at(-1) ?? 0 },
   shortTurnsUnder40Chars: S.humanTurns.filter((t) => t.text.trim().length < 40).length,
-  interrupts: S.interrupts,
+  interruptsMainSessions: S.interruptsMain,
+  interruptsAllRecords: S.interruptsAll,
   compactEvents: S.compacts,
   capacityLimitEvents: S.limits.length,
   capacityLimitSamples: S.limits.slice(0, 10),
   usage: S.usage,
   cacheReadToOutputRatio: S.usage.out ? +(S.usage.cacheRead / S.usage.out).toFixed(1) : null,
-  models: [...S.models].sort((a, b) => b[1].messages - a[1].messages),
+  models: [...S.models].sort((a, b) => b[1] - a[1]),
   topTools: [...S.tools].sort((a, b) => b[1] - a[1]).slice(0, 40),
   agentSpawns: [...S.agents].sort((a, b) => b[1] - a[1]),
   skillInvocations: [...S.skills].sort((a, b) => b[1] - a[1]),
@@ -233,6 +246,23 @@ const report = {
   sessions,
 };
 
+// Internal consistency gate. A report whose aggregates disagree with its own rows is not
+// evidence, and a previous version shipped exactly that. Fail loudly rather than emit it.
+const rowInterrupts = sessions.reduce((n, s) => n + s.interrupts, 0);
+const rowHuman = sessions.reduce((n, s) => n + s.human, 0);
+const problems = [];
+if (rowInterrupts !== report.interruptsMainSessions) {
+  problems.push(`interruptsMainSessions ${report.interruptsMainSessions} != sum of session rows ${rowInterrupts}`);
+}
+if (rowHuman !== report.humanTurnsOrganic) {
+  problems.push(`humanTurnsOrganic ${report.humanTurnsOrganic} != sum of session rows ${rowHuman}`);
+}
+if (problems.length) {
+  console.error('INCONSISTENT REPORT, refusing to write:\n  ' + problems.join('\n  '));
+  process.exit(3);
+}
+report.echoFilterVersion = 2;
+
 fs.writeFileSync(OUT, JSON.stringify(report, null, 1));
 fs.writeFileSync(OUT.replace(/\.json$/, '') + '-turns.json', JSON.stringify(S.humanTurns, null, 1));
 

package/adapters/cursor/olko-reflection/skills/retro-analysis/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: retro-analysis
-description: Produce a recurring engineering retrospective from repository history, delivery evidence, work patterns, code-quality signals, and prior snapshots, with repository, comparison, and cross-project modes.
+description: >
+  Produce a recurring engineering retrospective from repository history, delivery evidence, work
+  patterns, code-quality signals and prior snapshots, in repository, comparison or cross-project
+  mode. Use when the user asks for a retro over shipped work: "run a retrospective", "how did this
+  quarter go", "compare this month to last", "what did we ship", "engineering retro across my
+  repos", or when a scheduled retro job fires. This is the repository axis: for how the agents
+  themselves are being operated use agent-ops-retro, for a single finished change use wrap-up.
 license: MIT
 metadata:
   author: Oleg Koval

package/adapters/cursor/olko-reflection/skills/wrap-up/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: wrap-up
-description: Verify that a completed implementation matches its original objective, all applicable checks are green, and task-owned temporary files, worktrees, and local branches are safely tidied.
+description: >
+  Close out a finished piece of work: verify the implementation actually matches its original
+  objective, confirm every applicable check is green, and safely tidy the temporary files,
+  worktrees and local branches the task created. Use when a change is believed done and the user
+  says "wrap up", "we are done here", "finish up", "clean up after this", "tidy the worktrees", or
+  before handing work off. This is the single-change axis: for a retro over repository history use
+  retro-analysis, for how the agents are being operated use agent-ops-retro.
 license: MIT
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, Grok, and other Agent Skills compatible tools.
 metadata:

package/adapters/cursor/olko-release/skills/changelog-generator/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: changelog-generator
-description: Automatically creates user-facing changelogs from git commits by analyzing commit history, categorizing changes, and transforming technical commits into clear, customer-friendly release notes. Turns hours of manual changelog writing into minutes of automated generation.
+description: >
+  Turn git commit history into a user-facing changelog: read the commits since the last release,
+  categorise them, and rewrite technical messages as release notes a customer can actually read.
+  Use when the user asks for "a changelog", "release notes", "what changed since the last release",
+  "write the CHANGELOG entry", or is preparing a release and needs the notes written. For setting
+  up the release pipeline itself use semantic-release-beta.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires git history access in the target repository.

package/adapters/cursor/olko-release/skills/semantic-release-beta/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: semantic-release-beta
-description: Set up or update a Node.js release pipeline that uses semantic-release-npm-github-publish with a stable main channel and a prerelease beta channel on a beta branch, including GitHub Actions validation and npm publishing behavior.
+description: >
+  Set up or repair a Node.js release pipeline built on semantic-release-npm-github-publish, with a
+  stable channel on main and a prerelease channel on a beta branch, including the GitHub Actions
+  validation and npm publishing behaviour. Use when the user wants "semantic release set up", "a
+  beta release channel", "prerelease publishing", "automated npm publishing", or when an npm
+  package's releases are still manual and should become commit-driven. For writing the notes of a
+  single release use changelog-generator.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires a Node.js package repository that uses git and GitHub Actions.

package/adapters/cursor/olko-skill-meta/skills/ai-tools-setup/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-tools-setup
-description: 'Set up, repair, and report on the RTK+ICM+Vox AI development toolkit. Installs missing tools, fixes broken hooks and MCP config, and shows an effectiveness digest (token savings, memory accumulation, voice health) on demand or as a scheduled weekly report.'
+description: >
+  Set up, repair, and report on the RTK+ICM+Vox AI development toolkit: installs missing tools,
+  fixes broken hooks and MCP config, and shows an effectiveness digest covering token savings,
+  memory accumulation and voice health. Use when the user says "set up my AI tools", "my hooks are
+  broken", "MCP is not loading", "check my toolkit", "is ICM working", "how much are these tools
+  saving me", or when the scheduled weekly toolkit report fires.
 license: MIT
 allowed-tools: Bash, Read, Edit, Write
 compatibility: Claude Code on macOS (Homebrew required)

package/adapters/cursor/olko-skill-meta/skills/promptctl/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: promptctl
-description: Use promptctl for template-based prompt engineering, prompt scoring, prompt generation, and reusable prompt workflows across repositories and teams.
+description: >
+  Manage reusable prompts with the promptctl CLI: template-based prompt engineering, prompt
+  generation, and scoring a prompt before it ships. Use when the user wants to "write a reusable
+  prompt", "score this prompt", "turn this into a prompt template", "share prompts across repos",
+  or names promptctl directly. For bringing an agent skill into this catalog use
+  add-to-my-skills.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires promptctl to be installed when the workflow depends on the CLI.

package/adapters/cursor/olko-web-ops/skills/docs-index-keeper/SKILL.md

@@ -1,6 +1,10 @@
 ---
 name: docs-index-keeper
-description: Keep a repository's Markdown docs index in sync by installing and running docs-index-keeper in pre-commit hooks, CI, or one-off maintenance flows.
+description: >
+  Keep a repository's Markdown docs index in sync with the files that actually exist, by installing
+  and running docs-index-keeper in a pre-commit hook, in CI, or as a one-off sweep. Use when the
+  user says "the docs index is stale", "update the docs index", "add a docs index check to CI",
+  "our doc links are out of date", or when new docs were added without being linked anywhere.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit, Grep, Glob
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires shell access and Node.js 18+ when installing from npm.

package/adapters/cursor/olko-web-ops/skills/website-analytics-bootstrap/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: website-analytics-bootstrap
-description: Set up a website's first SEO analytics loop end-to-end: persistent SerpBear rank tracking, Google Search Console, seeded keywords, read-only technical audits, and Telegram alerts on a local host or NAS.
+description: >
+  Stand up a website's first SEO analytics loop end to end: persistent SerpBear rank tracking,
+  Google Search Console, seeded keywords, read-only technical audits and Telegram alerts on a local
+  host or NAS. Use when the user says "set up rank tracking", "this site has no analytics", "track
+  my keywords", "bootstrap SEO monitoring", or wants to know how a site performs in search when no
+  tracking exists yet. For diagnosing an existing indexing problem use
+  search-console-indexing-audit.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires network access, a reachable deployment host, and owner-provided Search Console and alert credentials.

package/adapters/grok/olko-creative/skills/vinted-listing/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: vinted-listing
-description: Create, edit, and safely publish Vinted listings through the signed-in browser, using verified item details and the owner’s original photos.
+description: >
+  Create, edit, and safely publish a Vinted listing through the signed-in browser, using verified
+  item details and the owner's original photos. Suggests title, description, category, brand,
+  condition, colour and parcel size from the photos, saves a draft first, and requires explicit
+  confirmation before publishing. Use when the user wants to list, sell, relist or price an item on
+  Vinted, or says "put this on Vinted", "create a Vinted listing", "sell this on Vinted", "edit my
+  Vinted draft", or "publish that listing". Browser workflow, not an API integration.
 ---
 <!-- Generated by scripts/build-adapters.sh. Do not edit directly. -->
 

package/adapters/grok/olko-git-tools/skills/gh-cli/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: gh-cli
-description: GitHub CLI (gh) comprehensive reference for repositories, issues, pull requests, Actions, projects, releases, gists, codespaces, organizations, extensions, and all GitHub operations from the command line.
+description: >
+  Reach for the GitHub CLI (gh) to do something on GitHub from the command line: repositories,
+  issues, pull requests, Actions runs and logs, projects, releases, gists, codespaces, org and
+  extension management. Use when a task needs a gh command and the exact syntax, flag or API path
+  is not obvious, or the user says "use gh", "what is the gh command for", "list my PRs from the
+  CLI", or "query the GitHub API". For driving a PR through review bots and CI use the
+  olko-github-pr skills instead; this one is the command reference.
 license: MIT
 allowed-tools: Bash
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires the GitHub CLI and valid GitHub authentication.

package/adapters/grok/olko-product/skills/viral-launch/SKILL.md

@@ -1,7 +1,11 @@
 ---
-
 name: viral-launch
-description: Set up a project repository and launch plan for shareable marketing, public launch readiness, and growth loops without pretending virality is guaranteed.
+description: >
+  Prepare a project for a public launch: shareable positioning, a launch-readiness pass over the
+  repository, and the growth loops worth building, without pretending virality is guaranteed. Use
+  when the user says "help me launch this", "prep for Product Hunt", "write the launch post", "make
+  this shareable", "launch plan", or is about to make a repo or product public. For building the
+  thing itself use mvp-oneshot or product-builder.
 license: MIT
 allowed-tools: Bash, Read, Write, WebSearch, WebFetch
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Useful for product repos, open-source packages, SaaS launches, waitlists, and creator tools.

package/adapters/grok/olko-reflection/skills/agent-ops-retro/scripts/mine-transcripts.mjs

@@ -27,7 +27,7 @@ const OUT = arg('out', './mine.json');
 const SINCE = Date.parse(`${arg('since', '')}T00:00:00Z`);
 const UNTIL = Date.parse(`${arg('until', '')}T00:00:00Z`);
 
-if (!Number.isFinite(SINCE) || !Number.isFinite(UNTIL) || UNTIL <= SINCE) {
+if (!Number.isFinite(SINCE) || !Number.isFinite(UNTIL)) {
   console.error('usage: mine-transcripts.mjs --since YYYY-MM-DD --until YYYY-MM-DD [--out path] [--root dir]');
   process.exit(2);
 }
@@ -46,7 +46,23 @@ function walk(dir, out = []) {
 }
 
 // A user-role text record that is really harness output, not something a person typed.
-const HARNESS = /^(<system-reminder|<command-name|<command-message|<local-command|<bash-input|<bash-stdout|<user-prompt-submit-hook|<task-notification|Caveat: The messages below|\[Request interrupted)/;
+// A user-role text record that is really harness output, not something a person typed.
+// Each alternative below was found by a run that mis-counted. The first pass caught the
+// obvious wrappers and still let 26% through: skill body dumps, image attachments,
+// already-loaded stubs and permission-grant echoes all sit in human-turn position.
+const HARNESS = new RegExp('^(' + [
+  '<system-reminder', '<command-name', '<command-message', '<local-command',
+  '<bash-input', '<bash-stdout', '<bash-stderr', '<user-prompt-submit-hook',
+  '<task-notification', '<cross-session-message',
+  'Caveat: The messages below',
+  '\\[Request interrupted',
+  'Base directory for this skill:',      // a skill body dumped into the turn
+  'Skill /[^ ]+ is already loaded',      // re-invocation stub
+  'Permission granted for:',             // permission-grant echo
+  '\\[Image #\\d+\\]',                    // bare attachment
+  '\\[Image: source:',
+  '\\[Pasted text',
+].join('|') + ')');
 
 const textOf = (msg) => {
   const c = msg?.content;
@@ -68,7 +84,8 @@ const S = {
   limits: [],
   humanTurns: [],
   echoTurns: 0,
-  interrupts: 0,
+  interruptsAll: 0,
+  interruptsMain: 0,
   compacts: 0,
   usage: { in: 0, out: 0, cacheRead: 0, cacheWrite: 0 },
 };
@@ -134,16 +151,8 @@ for (const f of files) {
 
     if (r.type === 'assistant') {
       const model = r.message?.model;
+      if (model) bump(S.models, model);
       const u = r.message?.usage;
-      if (model) {
-        const stats = S.models.get(model) || { messages: 0, in: 0, out: 0, cacheRead: 0, cacheWrite: 0 };
-        stats.messages++;
-        stats.in += u?.input_tokens || 0;
-        stats.out += u?.output_tokens || 0;
-        stats.cacheRead += u?.cache_read_input_tokens || 0;
-        stats.cacheWrite += u?.cache_creation_input_tokens || 0;
-        S.models.set(model, stats);
-      }
       if (u) {
         S.usage.in += u.input_tokens || 0;
         S.usage.out += u.output_tokens || 0;
@@ -182,9 +191,12 @@ for (const f of files) {
     }
 
     const raw = textOf(r.message);
+    // Two scopes, two names. The first version reported the all-records count and the
+    // per-session count under one label `interrupts`, so the top-level figure (27) did not
+    // match the sum of the per-session rows (15) and neither number could be trusted.
     if (raw.includes('[Request interrupted')) {
-      S.interrupts++;
-      if (sess) sess.interrupts++;
+      S.interruptsAll++;
+      if (sess) { S.interruptsMain++; sess.interrupts++; }
     }
     if (!main || r.userType !== 'external' || !raw || sawToolResult) continue;
 
@@ -219,13 +231,14 @@ const report = {
   humanTurnsPerDay: byDay,
   turnsPerSession: { median: pct(0.5), p90: pct(0.9), max: humanCounts.at(-1) ?? 0 },
   shortTurnsUnder40Chars: S.humanTurns.filter((t) => t.text.trim().length < 40).length,
-  interrupts: S.interrupts,
+  interruptsMainSessions: S.interruptsMain,
+  interruptsAllRecords: S.interruptsAll,
   compactEvents: S.compacts,
   capacityLimitEvents: S.limits.length,
   capacityLimitSamples: S.limits.slice(0, 10),
   usage: S.usage,
   cacheReadToOutputRatio: S.usage.out ? +(S.usage.cacheRead / S.usage.out).toFixed(1) : null,
-  models: [...S.models].sort((a, b) => b[1].messages - a[1].messages),
+  models: [...S.models].sort((a, b) => b[1] - a[1]),
   topTools: [...S.tools].sort((a, b) => b[1] - a[1]).slice(0, 40),
   agentSpawns: [...S.agents].sort((a, b) => b[1] - a[1]),
   skillInvocations: [...S.skills].sort((a, b) => b[1] - a[1]),
@@ -233,6 +246,23 @@ const report = {
   sessions,
 };
 
+// Internal consistency gate. A report whose aggregates disagree with its own rows is not
+// evidence, and a previous version shipped exactly that. Fail loudly rather than emit it.
+const rowInterrupts = sessions.reduce((n, s) => n + s.interrupts, 0);
+const rowHuman = sessions.reduce((n, s) => n + s.human, 0);
+const problems = [];
+if (rowInterrupts !== report.interruptsMainSessions) {
+  problems.push(`interruptsMainSessions ${report.interruptsMainSessions} != sum of session rows ${rowInterrupts}`);
+}
+if (rowHuman !== report.humanTurnsOrganic) {
+  problems.push(`humanTurnsOrganic ${report.humanTurnsOrganic} != sum of session rows ${rowHuman}`);
+}
+if (problems.length) {
+  console.error('INCONSISTENT REPORT, refusing to write:\n  ' + problems.join('\n  '));
+  process.exit(3);
+}
+report.echoFilterVersion = 2;
+
 fs.writeFileSync(OUT, JSON.stringify(report, null, 1));
 fs.writeFileSync(OUT.replace(/\.json$/, '') + '-turns.json', JSON.stringify(S.humanTurns, null, 1));
 

package/adapters/grok/olko-reflection/skills/retro-analysis/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: retro-analysis
-description: Produce a recurring engineering retrospective from repository history, delivery evidence, work patterns, code-quality signals, and prior snapshots, with repository, comparison, and cross-project modes.
+description: >
+  Produce a recurring engineering retrospective from repository history, delivery evidence, work
+  patterns, code-quality signals and prior snapshots, in repository, comparison or cross-project
+  mode. Use when the user asks for a retro over shipped work: "run a retrospective", "how did this
+  quarter go", "compare this month to last", "what did we ship", "engineering retro across my
+  repos", or when a scheduled retro job fires. This is the repository axis: for how the agents
+  themselves are being operated use agent-ops-retro, for a single finished change use wrap-up.
 license: MIT
 metadata:
   author: Oleg Koval

package/adapters/grok/olko-reflection/skills/wrap-up/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: wrap-up
-description: Verify that a completed implementation matches its original objective, all applicable checks are green, and task-owned temporary files, worktrees, and local branches are safely tidied.
+description: >
+  Close out a finished piece of work: verify the implementation actually matches its original
+  objective, confirm every applicable check is green, and safely tidy the temporary files,
+  worktrees and local branches the task created. Use when a change is believed done and the user
+  says "wrap up", "we are done here", "finish up", "clean up after this", "tidy the worktrees", or
+  before handing work off. This is the single-change axis: for a retro over repository history use
+  retro-analysis, for how the agents are being operated use agent-ops-retro.
 license: MIT
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, Grok, and other Agent Skills compatible tools.
 metadata:

package/adapters/grok/olko-release/skills/changelog-generator/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: changelog-generator
-description: Automatically creates user-facing changelogs from git commits by analyzing commit history, categorizing changes, and transforming technical commits into clear, customer-friendly release notes. Turns hours of manual changelog writing into minutes of automated generation.
+description: >
+  Turn git commit history into a user-facing changelog: read the commits since the last release,
+  categorise them, and rewrite technical messages as release notes a customer can actually read.
+  Use when the user asks for "a changelog", "release notes", "what changed since the last release",
+  "write the CHANGELOG entry", or is preparing a release and needs the notes written. For setting
+  up the release pipeline itself use semantic-release-beta.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires git history access in the target repository.

package/adapters/grok/olko-release/skills/semantic-release-beta/SKILL.md

@@ -1,6 +1,12 @@
 ---
 name: semantic-release-beta
-description: Set up or update a Node.js release pipeline that uses semantic-release-npm-github-publish with a stable main channel and a prerelease beta channel on a beta branch, including GitHub Actions validation and npm publishing behavior.
+description: >
+  Set up or repair a Node.js release pipeline built on semantic-release-npm-github-publish, with a
+  stable channel on main and a prerelease channel on a beta branch, including the GitHub Actions
+  validation and npm publishing behaviour. Use when the user wants "semantic release set up", "a
+  beta release channel", "prerelease publishing", "automated npm publishing", or when an npm
+  package's releases are still manual and should become commit-driven. For writing the notes of a
+  single release use changelog-generator.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires a Node.js package repository that uses git and GitHub Actions.

package/adapters/grok/olko-skill-meta/skills/ai-tools-setup/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: ai-tools-setup
-description: 'Set up, repair, and report on the RTK+ICM+Vox AI development toolkit. Installs missing tools, fixes broken hooks and MCP config, and shows an effectiveness digest (token savings, memory accumulation, voice health) on demand or as a scheduled weekly report.'
+description: >
+  Set up, repair, and report on the RTK+ICM+Vox AI development toolkit: installs missing tools,
+  fixes broken hooks and MCP config, and shows an effectiveness digest covering token savings,
+  memory accumulation and voice health. Use when the user says "set up my AI tools", "my hooks are
+  broken", "MCP is not loading", "check my toolkit", "is ICM working", "how much are these tools
+  saving me", or when the scheduled weekly toolkit report fires.
 license: MIT
 allowed-tools: Bash, Read, Edit, Write
 compatibility: Claude Code on macOS (Homebrew required)

package/adapters/grok/olko-skill-meta/skills/promptctl/SKILL.md

@@ -1,6 +1,11 @@
 ---
 name: promptctl
-description: Use promptctl for template-based prompt engineering, prompt scoring, prompt generation, and reusable prompt workflows across repositories and teams.
+description: >
+  Manage reusable prompts with the promptctl CLI: template-based prompt engineering, prompt
+  generation, and scoring a prompt before it ships. Use when the user wants to "write a reusable
+  prompt", "score this prompt", "turn this into a prompt template", "share prompts across repos",
+  or names promptctl directly. For bringing an agent skill into this catalog use
+  add-to-my-skills.
 license: MIT
 allowed-tools: Bash, Read, Write
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires promptctl to be installed when the workflow depends on the CLI.

package/adapters/grok/olko-web-ops/skills/docs-index-keeper/SKILL.md

@@ -1,6 +1,10 @@
 ---
 name: docs-index-keeper
-description: Keep a repository's Markdown docs index in sync by installing and running docs-index-keeper in pre-commit hooks, CI, or one-off maintenance flows.
+description: >
+  Keep a repository's Markdown docs index in sync with the files that actually exist, by installing
+  and running docs-index-keeper in a pre-commit hook, in CI, or as a one-off sweep. Use when the
+  user says "the docs index is stale", "update the docs index", "add a docs index check to CI",
+  "our doc links are out of date", or when new docs were added without being linked anywhere.
 license: MIT
 allowed-tools: Bash, Read, Write, Edit, Grep, Glob
 compatibility: Codex, Claude Code, Cursor, GitHub Copilot, Windsurf, Kiro, and other Agent Skills compatible tools. Requires shell access and Node.js 18+ when installing from npm.