@saifymatteo/pi-tool-usage-log 0.0.0-stage → 0.1.1

package/src/skill-usage-log.ts

@@ -0,0 +1,316 @@
+/**
+ * skill-usage-log.ts — track skill *use*, not just reads, so unused skills
+ * can be identified for removal.
+ *
+ * A skill is considered USED when a successful SKILL.md read is followed by
+ * another tool call while answering the same request, even in a later model
+ * turn. This is an activity heuristic, not proof the instructions were followed.
+ * A skill that is read then abandoned (no follow-up tool) is NOT counted.
+ *
+ * Emits two event kinds to <agent-home>/skill-usage.log
+ * (defaults to ~/.pi/agent, or PI_CODING_AGENT_DIR):
+ *   skill:load      — SKILL.md read (deduped once per session)
+ *   skill:complete  — skill loaded AND had follow-up tool activity this request
+ * Both events are deduped once per session. New TSV rows append the selected
+ * agent model and thinking level to the legacy six fields. Timestamps use the
+ * runtime's local timezone with explicit offsets; old UTC rows remain readable.
+ *
+ * /skill-usage summarizes the log across projects, sorted by used-session count.
+ * It lists every installed skill (pi.getCommands() skill entries), so never-loaded skills show as 0/never.
+ * Historical rows retain the old counting heuristic; they cannot be repaired.
+ * `buildTracker()` is pure for testing; the default export wires it up to pi.
+ */
+import type { ExtensionAPI, ExtensionContext, Theme } from '@earendil-works/pi-coding-agent'
+import { getMarkdownTheme } from '@earendil-works/pi-coding-agent'
+import { Markdown, matchesKey, truncateToWidth, visibleWidth, type Component, type TUI, type TuiMouseEvent } from '@earendil-works/pi-tui'
+import { appendFileSync, mkdirSync } from 'node:fs'
+import { readFile } from 'node:fs/promises'
+import { homedir } from 'node:os'
+import { dirname, join, resolve } from 'node:path'
+import { agentMetadata, localTimestamp, logField, markdownCell, usageTimeFormat } from './lib/usage-log.ts'
+import { DIALOG_OVERLAY, dialogRows, wheelAsArrows } from './lib/dialog.ts'
+
+export interface TrackerEvent {
+  event: 'skill:load' | 'skill:complete'
+  skill: string
+  path: string
+  model?: string
+  thinking?: string
+}
+
+export interface SkillTracker {
+  handleRequestStart(): void
+  handleToolCall(tool: string | undefined, metadata?: { model: string; thinking: string }): void
+  handleSkillRead(path: string, metadata?: { model: string; thinking: string }): void
+  handleTurnEnd(): TrackerEvent[]
+}
+
+/** Pure skill-usage state machine — no pi, no fs. Testable. */
+export function buildTracker(onEvent?: (e: TrackerEvent) => void): SkillTracker {
+  const sessionUsed = new Set<string>()
+  const sessionLoaded = new Set<string>()
+  const requestLoaded = new Map<string, string>()
+  const requestUsed = new Map<string, { model: string; thinking: string } | undefined>()
+
+  return {
+    handleRequestStart() {
+      requestLoaded.clear()
+      requestUsed.clear()
+    },
+    handleToolCall(tool, metadata) {
+      if (tool && tool !== 'think') {
+        for (const skill of requestLoaded.keys()) {
+          if (!requestUsed.has(skill)) requestUsed.set(skill, metadata && { ...metadata })
+        }
+      }
+    },
+    handleSkillRead(path, metadata) {
+      const norm = normalize(path)
+      // Same SKILL.md rule as compact-tools; keep legacy standalone skills too.
+      const skill = norm.match(/(?:^|\/)([^/]+)\/SKILL\.md$/i)?.[1] ??
+        norm.match(/(?:^|\/)skills\/([^/]+)\.md$/i)?.[1]
+      if (!skill) return
+      requestLoaded.set(skill, path)
+      if (!sessionLoaded.has(skill)) {
+        sessionLoaded.add(skill)
+        onEvent?.({ event: 'skill:load', skill, path, ...metadata })
+      }
+    },
+    handleTurnEnd() {
+      const out: TrackerEvent[] = []
+      for (const [skill, metadata] of requestUsed) {
+        if (!sessionUsed.has(skill)) {
+          sessionUsed.add(skill)
+          out.push({ event: 'skill:complete', skill, path: requestLoaded.get(skill)!, ...metadata })
+        }
+      }
+      out.forEach((event) => onEvent?.(event))
+      return out
+    },
+  }
+}
+
+function normalize(p: string): string {
+  return p.replaceAll('\\', '/')
+}
+
+function isModelInvocationDisabled(content: string): boolean {
+  const frontmatter = content.match(/^---\s*\r?\n([\s\S]*?)\r?\n---(?:\r?\n|$)/)?.[1]
+  return frontmatter?.split(/\r?\n/).some((line) => /^disable-model-invocation:\s*true\s*(?:#.*)?$/i.test(line)) ?? false
+}
+
+export interface UsageRow {
+  skill: string
+  loads: number
+  uses: number
+  lastUsed: string | null
+}
+
+/** Counts sessions, not tool calls; accepts legacy six- and metadata eight-column TSV.
+ *  When supplied, `installed` filters removed skills and adds unseen skills as zero rows. */
+export function summarizeUsage(log: string, installed?: string[]): { skills: UsageRow[]; skipped: number } {
+  const { skills, skipped } = summarizeLog(log)
+  if (installed === undefined) return { skills, skipped }
+  const available = new Set(installed)
+  const filtered = skills.filter((row) => available.has(row.skill))
+  const seen = new Set(filtered.map((row) => row.skill))
+  const unused = [...available].filter((skill) => !seen.has(skill)).sort((a, b) => a.localeCompare(b))
+  return { skills: [...filtered, ...unused.map((skill) => ({ skill, loads: 0, uses: 0, lastUsed: null }))], skipped }
+}
+
+function summarizeLog(log: string, byAgent = false) {
+  const rows = new Map<string, { row: UsageRow & { model?: string; thinking?: string }; loads: Set<string>; uses: Set<string> }>()
+  let skipped = 0
+  for (const line of log.split(/\r?\n/)) {
+    if (!line) continue
+    const fields = line.split('\t')
+    const [timestamp, sid, , event, skill] = fields
+    const time = Date.parse(timestamp)
+    if ((fields.length !== 6 && fields.length !== 8) || !Number.isFinite(time) || !skill ||
+        /[\x00-\x1f\x7f-\x9f]/.test(skill) ||
+        (event !== 'skill:load' && event !== 'skill:complete')) {
+      skipped++
+      continue
+    }
+    const model = fields[6] || 'unknown'
+    const thinking = fields[7] || 'unknown'
+    const group = byAgent ? JSON.stringify([skill, model, thinking]) : skill
+    let entry = rows.get(group)
+    if (!entry) {
+      entry = { row: { skill, loads: 0, uses: 0, lastUsed: null, ...(byAgent ? { model, thinking } : {}) }, loads: new Set(), uses: new Set() }
+      rows.set(group, entry)
+    }
+    // Old logs can have an empty sid; do not merge their unrelated uses.
+    const key = sid || timestamp
+    if (event === 'skill:load') {
+      entry.loads.add(key)
+      entry.row.loads = entry.loads.size
+    } else {
+      entry.uses.add(key)
+      entry.row.uses = entry.uses.size
+      const usedAt = new Date(time).toISOString()
+      if (!entry.row.lastUsed || usedAt > entry.row.lastUsed) entry.row.lastUsed = usedAt
+    }
+  }
+  return {
+    skills: [...rows.values()].map(({ row }) => row)
+      .sort((a, b) => b.uses - a.uses || a.skill.localeCompare(b.skill)),
+    skipped,
+  }
+}
+
+/** UI-only viewport: opening/closing it never appends session content. */
+export class SkillUsageDialog implements Component {
+  private offset = 0
+  private pageSize = 1
+  private lineCount = 0
+  private readonly document: Markdown
+
+  constructor(private readonly tui: TUI, private readonly theme: Theme, report: string, private readonly done: () => void, private readonly title = 'Skill usage') {
+    this.document = new Markdown(report, 0, 0, getMarkdownTheme())
+  }
+
+  invalidate(): void { this.document.invalidate() }
+
+  render(width: number): string[] {
+    if (width < 1) return []
+    const bodyWidth = Math.max(1, width - 4)
+    const content = this.document.render(bodyWidth)
+    this.lineCount = content.length
+    // Keep the footer visible on short terminals, including after a resize.
+    // Fixed lines: 2 borders, footer.
+    this.pageSize = dialogRows(this.tui, 3)
+    this.offset = Math.max(0, Math.min(this.offset, this.lineCount - this.pageSize))
+    const border = (text: string) => this.theme.fg('border', text)
+    const row = (text: string) => {
+      const fitted = truncateToWidth(text, bodyWidth, '…')
+      return border('┃') + ` ${fitted}${' '.repeat(Math.max(0, bodyWidth - visibleWidth(fitted)))} ` + border('┃')
+    }
+    const title = truncateToWidth(` ${this.title} `, Math.max(0, width - 2), '…')
+    const range = `${this.lineCount ? this.offset + 1 : 0}–${Math.min(this.offset + this.pageSize, this.lineCount)}/${this.lineCount}`
+    const lines = [border(`┏${title}${'━'.repeat(Math.max(0, width - 2 - visibleWidth(title)))}┓`),
+      ...Array.from({ length: this.pageSize }, (_, i) => row(content[this.offset + i] ?? '')),
+      row(this.theme.fg('dim', `Esc close · ↑↓ scroll · PgUp/PgDn · ${range}`)),
+      border(`┗${'━'.repeat(Math.max(0, width - 2))}┛`)]
+    return lines.map((line) => truncateToWidth(line, width, ''))
+  }
+
+  handleMouse(event: TuiMouseEvent) { return wheelAsArrows(event, (key) => this.handleInput(key)) }
+
+  handleInput(data: string): void {
+    if (matchesKey(data, 'escape') || matchesKey(data, 'ctrl+c') || matchesKey(data, 'enter')) return this.done()
+    if (matchesKey(data, 'up')) this.offset--
+    else if (matchesKey(data, 'down')) this.offset++
+    else if (matchesKey(data, 'pageUp')) this.offset -= this.pageSize
+    else if (matchesKey(data, 'pageDown')) this.offset += this.pageSize
+    else if (matchesKey(data, 'home')) this.offset = 0
+    else if (matchesKey(data, 'end')) this.offset = this.lineCount
+    else return
+    this.offset = Math.max(0, Math.min(this.offset, this.lineCount - this.pageSize))
+    this.tui.requestRender()
+  }
+}
+
+export default function register(pi: ExtensionAPI): void {
+  const agentDir = process.env.PI_CODING_AGENT_DIR || join(homedir(), '.pi', 'agent')
+  const logPath = join(agentDir.replace(/^~(?=$|[\\/])/, homedir()), 'skill-usage.log')
+  let currentSid: string | undefined
+  let currentCwd = ''
+  let tracker: SkillTracker
+  let logError: string | undefined
+  let warned = false
+
+  function forSession(ctx: ExtensionContext): SkillTracker {
+    const sid = ctx.sessionManager.getSessionId()
+    currentCwd = ctx.cwd
+    if (sid !== currentSid || !tracker) {
+      currentSid = sid
+      warned = false
+      logError = undefined
+      tracker = buildTracker((e) => {
+        try {
+          mkdirSync(dirname(logPath), { recursive: true })
+          const fields = [localTimestamp(), sid, currentCwd, e.event, e.skill, normalize(e.path), e.model ?? 'unknown', e.thinking ?? 'unknown']
+          // Keep unusual filenames and working directories from corrupting TSV rows.
+          appendFileSync(logPath, fields.map(logField).join('\t') + '\n')
+        } catch (error) {
+          logError = error instanceof Error ? error.message : String(error)
+        }
+      })
+    }
+    return tracker
+  }
+
+  pi.on('session_start', (_event, ctx) => { forSession(ctx) })
+  pi.on('before_agent_start', (_event, ctx) => { forSession(ctx).handleRequestStart() })
+  pi.on('tool_call', (event, ctx) => { forSession(ctx).handleToolCall(event.toolName, agentMetadata(pi, ctx)) })
+  pi.on('tool_result', (event, ctx) => {
+    if (event.toolName !== 'read' || event.isError || typeof event.input.path !== 'string') return
+    const raw = event.input.path.replace(/^~(?=$|[\\/])/, homedir())
+    forSession(ctx).handleSkillRead(resolve(ctx.cwd, raw), agentMetadata(pi, ctx))
+  })
+  pi.on('turn_end', (_event, ctx) => {
+    forSession(ctx).handleTurnEnd()
+    if (logError && !warned && ctx.hasUI) {
+      warned = true
+      ctx.ui.notify(`Skill usage could not be logged to ${logPath}: ${logError}`, 'warning')
+    }
+  })
+
+  pi.registerCommand('skill-usage', {
+    description: 'List historical skill usage, agent models, thinking levels and local last-used times',
+    handler: async (args, ctx) => {
+      if (ctx.mode !== 'tui') {
+        ctx.ui.notify('Skill usage requires interactive terminal mode; no report was added to the transcript.', 'info')
+        return
+      }
+      if (args.trim()) {
+        ctx.ui.notify('Usage: /skill-usage (all logged sessions and projects)', 'info')
+        return
+      }
+      try {
+        let log = ''
+        try {
+          log = await readFile(logPath, 'utf8')
+        } catch (error) {
+          if ((error as NodeJS.ErrnoException).code !== 'ENOENT') throw error
+        }
+        const skillCommands = pi.getCommands().filter((command) => command.source === 'skill')
+        const installed = skillCommands.map((command) => command.name.replace(/^skill:/, ''))
+        const invocationDisabled = new Map<string, boolean>()
+        await Promise.all(skillCommands.map(async (command, index) => {
+          const path = command.sourceInfo?.path
+          if (typeof path !== 'string') return
+          try { invocationDisabled.set(installed[index], isModelInvocationDisabled(await readFile(path, 'utf8'))) }
+          catch { /* Missing/unreadable skill metadata is reported as invocation-enabled. */ }
+        }))
+        const { skills, skipped } = summarizeUsage(log, installed)
+        const time = usageTimeFormat()
+        const lines = ['Skill usage — all logged sessions and projects', '',
+          'Counts are unique sessions, not tool calls. Used = read followed by tool activity in the same request.',
+          'Older entries used a same-model-turn rule and may undercount. User invoke shows disable-model-invocation from skill frontmatter; 0 / never = not used since logging began.', '',
+          `| Skill | User invoke | Loaded sessions | Used sessions | Last used (${markdownCell(time.label)}) |`,
+          '| --- | --- | ---: | ---: | --- |']
+        for (const row of skills) {
+          lines.push(`| ${markdownCell(row.skill)} | ${invocationDisabled.get(row.skill) ? 'Yes' : '-'} | ${row.loads} | ${row.uses} | ${row.lastUsed ? markdownCell(time.format(new Date(row.lastUsed))) : 'never'} |`)
+        }
+        if (skills.some((row) => row.loads)) {
+          lines.push('', 'By selected agent model and thinking level (unique sessions within each group; groups may overlap):',
+            'Legacy metadata is unknown. Only the first load/use per session is logged; this is not every model that later used the skill.', '',
+            '| Skill | Model | Thinking | Loaded sessions | Used sessions |',
+            '| --- | --- | --- | ---: | ---: |')
+          for (const row of summarizeLog(log, true).skills.filter((row) => installed.includes(row.skill))) {
+            lines.push(`| ${markdownCell(row.skill)} | ${markdownCell(row.model!)} | ${markdownCell(row.thinking!)} | ${row.loads} | ${row.uses} |`)
+          }
+        }
+        if (!skills.some((row) => row.loads)) lines.push('', 'No skill usage recorded yet.')
+        if (skipped) lines.push('', `Skipped ${skipped} malformed log row(s).`)
+        lines.push('', `Log: ${normalize(logPath).replace(/[\x00-\x1f\x7f-\x9f]/g, ' ')}`)
+        await ctx.ui.custom<void>((tui, theme, _keybindings, done) =>
+          new SkillUsageDialog(tui, theme, lines.join('\n'), done), DIALOG_OVERLAY)
+      } catch (error) {
+        ctx.ui.notify(`Could not show skill usage: ${error instanceof Error ? error.message : String(error)}`, 'error')
+      }
+    },
+  })
+}