v0.1.0

First release of agentcov, which records the lines of a repository that AI coding
agents actually read and renders them as coverage.

• agentcov report writes LCOV tracefiles, textual .gcov files, and native JSON
  with per-line read counts, agentcov html writes a self-contained heatmap, and
  agentcov summary prints repository-wide read coverage while agentcov unread
  lists each file's unread lines and ranges. --counts full writes observed read
  counts into LCOV and gcov output; --counts binary writes read/unread.
• agentcov install-codex-hooks installs Codex hooks for the user or a single
  repository, capturing reads live from Bash, apply_patch, and MCP tools, and
  agentcov uninstall-codex-hooks removes them again.
• agentcov backfill replays an existing Codex, Claude Code, or Pi session by
  session id or transcript path, with --agent auto detecting which. Pi child
  sessions linked by parentSession are included.
• agentcov import reads ranges recorded by another tool, either
  agent-coverage-style JSON or a plain array of command and range entries.
• Reads are recognized from sed, head, tail, cat, awk, sort, and
  cat/nl piped into sed, including inside command lines joined with &&
  or ; and inside simple for loops. rg and grep output is recorded
  separately as search_seen, with match lines and context lines counted apart
  from each other.
• Command shapes that cannot be resolved to a line range are recorded as unknown
  events with a reason rather than counted as whole-file reads.
• Git-tracked text files that were never read are reported at zero coverage, so
  untouched files remain visible.
• Reports carry attribution for each read: agent, session, tool, command, task,
  timestamp, and confidence.
• .agentcov.toml configures the storage directory, exclusions, lockfile handling,
  and which reports are refreshed automatically on Codex Stop.
• Published to PyPI as agentcov, requiring Python 3.11 or newer and no runtime
  dependencies.