verifaied 0.25.0.dev53__tar.gz → 0.28.0.dev71__tar.gz

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/.gitignore

@@ -69,12 +69,18 @@ TODO.md
 *.tsbuildinfo
 
 # Parallel worktrees. `.worktree.mk` is the per-checkout port/project block
-# the setup-worktree skill generates (`.env.worktree`, its shell twin, is
+# `.agent/worktree-setup.sh` generates (`.env.worktree`, its shell twin, is
 # already covered by `**/.env.*` above). `.mcp.json` carries a live
 # `vr_live_…` bearer token — it is machine-local config, never repo config.
 .worktree.mk
 .mcp.json
 
+# The coordinator's marker for the tree it is standing in: which slot it
+# claims, and whether its build ever finished. Written per checkout, read
+# by the daemon, meaningless anywhere else. Everything else in `.agent/` is
+# committed on purpose.
+.agent/worktree.json
+
 # Beads / Dolt files (added by bd init)
 .dolt/
 *.db

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: verifaied
-Version: 0.25.0.dev53
+Version: 0.28.0.dev71
 Summary: Find what's untested in your code — locally, no account required
 Project-URL: Homepage, https://pypi.org/project/verifaied/
 Author: Kyle Richards
@@ -425,6 +425,11 @@ verifaied test run --all
 # Point any of them at another configured environment.
 verifaied test run --all --env stage
 
+# How much of the run's clock goes on making the video watchable.
+# `fast` (the default) marks what each step clicks and moves on;
+# `watch` lingers on every action; `off` records without annotations.
+verifaied test run "Settings/API tokens" --video watch
+
 # What's recorded, and how each last went on this branch.
 verifaied test list
 
@@ -918,11 +923,18 @@ it — the CLI finds the first one that has it, and `--runner-dir`
 overrides). This is Node Playwright, not the Python `verifaied[audit]`
 extra.
 
-**`test record` and `test studio` only run against Playwright 1.60.x.** Recording drives
-Playwright's recorder through an internal API — the same entry point
-`playwright codegen` uses — so it refuses any version it hasn't been
-verified against rather than opening a browser that might record nothing.
-`test run` uses only public API and works on any version.
+**`test record` and `test studio` need `@playwright/test` 1.60 or later.**
+Recording is driven by our own Node program through public Playwright API
+— the selector generator behind `PWDEBUG=console`, the AI-mode
+accessibility snapshot and the `aria-ref=` selectors its refs resolve
+through — and every recording ends in a replay, whose video marks what
+was clicked with `video.show`. 1.60 is where the last of those shipped,
+so it is a floor and not an allowlist: anything newer is allowed, and
+anything older is refused rather than opening a browser that would
+record nothing. `test run` stays ungated. Below 1.60 the annotation is
+not available, so set `replay_show_actions_ms = 0` and
+`replay_video_step_banner = false` there — that writes exactly the
+config the CLI wrote before this option existed.
 
 ### Exit codes
 
@@ -961,12 +973,46 @@ and `--no-replay` to skip the replay. `run` takes `--all` in place of a
 test name. `studio` takes `--no-replay` and takes no test name at all —
 what it records is whatever the web asks for.
 
+## `verifaied skill install` — hand the loop to your agent
+
+The CLI ships the agent skill for using it. One command writes it where a
+coding agent will find it:
+
+```bash
+verifaied skill install          # this project: .claude/skills/verifaied
+verifaied skill install --user   # every project: ~/.claude/skills/verifaied
+```
+
+The skill teaches the loop below — run the tests with coverage, `verifaied
+upload`, `verifaied check-done`, fix what comes back, repeat until done — plus
+the vitest/Istanbul requirement for TypeScript, the recorder for browser
+flows, `verifaied audit`, and the rules that keep the loop honest (no lowered
+thresholds, no `# pragma: no cover`, no `xfail` over a real failure).
+
+`install` with no flags writes into the **repository root** you are standing
+in, so running it from a subdirectory still lands the skill where your agent
+looks for it. Outside a repository it writes into the current directory.
+
+Because it ships inside this package, the skill can never describe a command
+the installed CLI does not have. It is two files, `SKILL.md` and a command
+reference the agent opens only when it needs a flag, and they land in a
+`verifaied` folder inside the skills directory — yours to read and, in the
+project-local case, to commit for the rest of the team.
+
+Re-running the command writes nothing when the installed copy already matches.
+A copy that differs, because you edited it or because you have since upgraded
+the CLI, is left alone until you pass `--force`, which replaces the files
+verifAIed ships and never removes anything you added alongside them. Install
+somewhere else entirely with `--dir <path>`.
+
+Exit codes: `0` installed, updated, or already up to date; `1` usage error
+(`--user` with `--dir`, or a differing copy without `--force`).
+
 ## Agent loop
 
-We've found the most fruitful way to use verifAIed is to put a short
-loop into your coding agent's instructions file (`CLAUDE.md`,
-`AGENTS.md`, `.cursorrules`, `GEMINI.md`,
-`.github/copilot-instructions.md`) so it self-corrects on every change.
+Agents that do not read `.claude/skills/` want the same loop in their
+instructions file (`CLAUDE.md`, `AGENTS.md`, `.cursorrules`, `GEMINI.md`,
+`.github/copilot-instructions.md`), so they self-correct on every change.
 
 The easiest way is to let your agent write that loop into the rules
 file for you. Paste the prompt below into your agent — it will inspect
@@ -1046,7 +1092,7 @@ After implementing or modifying any code in this repo, run this loop until verif
    ```
 2. Upload the results to verifAIed:
    ```
-   verifaied upload
+   verifaied upload --junit junit.xml
    ```
 3. Ask if you're done: call the `check_done` tool on the verifAIed MCP server (or run `verifaied check-done`). It returns `{done, status, gaps}` for this branch — failing tests plus untested/partially-covered functions among the ones you changed.
 4. If `done` is true, stop — the branch is covered.

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/README.md

@@ -379,6 +379,11 @@ verifaied test run --all
 # Point any of them at another configured environment.
 verifaied test run --all --env stage
 
+# How much of the run's clock goes on making the video watchable.
+# `fast` (the default) marks what each step clicks and moves on;
+# `watch` lingers on every action; `off` records without annotations.
+verifaied test run "Settings/API tokens" --video watch
+
 # What's recorded, and how each last went on this branch.
 verifaied test list
 
@@ -872,11 +877,18 @@ it — the CLI finds the first one that has it, and `--runner-dir`
 overrides). This is Node Playwright, not the Python `verifaied[audit]`
 extra.
 
-**`test record` and `test studio` only run against Playwright 1.60.x.** Recording drives
-Playwright's recorder through an internal API — the same entry point
-`playwright codegen` uses — so it refuses any version it hasn't been
-verified against rather than opening a browser that might record nothing.
-`test run` uses only public API and works on any version.
+**`test record` and `test studio` need `@playwright/test` 1.60 or later.**
+Recording is driven by our own Node program through public Playwright API
+— the selector generator behind `PWDEBUG=console`, the AI-mode
+accessibility snapshot and the `aria-ref=` selectors its refs resolve
+through — and every recording ends in a replay, whose video marks what
+was clicked with `video.show`. 1.60 is where the last of those shipped,
+so it is a floor and not an allowlist: anything newer is allowed, and
+anything older is refused rather than opening a browser that would
+record nothing. `test run` stays ungated. Below 1.60 the annotation is
+not available, so set `replay_show_actions_ms = 0` and
+`replay_video_step_banner = false` there — that writes exactly the
+config the CLI wrote before this option existed.
 
 ### Exit codes
 
@@ -915,12 +927,46 @@ and `--no-replay` to skip the replay. `run` takes `--all` in place of a
 test name. `studio` takes `--no-replay` and takes no test name at all —
 what it records is whatever the web asks for.
 
+## `verifaied skill install` — hand the loop to your agent
+
+The CLI ships the agent skill for using it. One command writes it where a
+coding agent will find it:
+
+```bash
+verifaied skill install          # this project: .claude/skills/verifaied
+verifaied skill install --user   # every project: ~/.claude/skills/verifaied
+```
+
+The skill teaches the loop below — run the tests with coverage, `verifaied
+upload`, `verifaied check-done`, fix what comes back, repeat until done — plus
+the vitest/Istanbul requirement for TypeScript, the recorder for browser
+flows, `verifaied audit`, and the rules that keep the loop honest (no lowered
+thresholds, no `# pragma: no cover`, no `xfail` over a real failure).
+
+`install` with no flags writes into the **repository root** you are standing
+in, so running it from a subdirectory still lands the skill where your agent
+looks for it. Outside a repository it writes into the current directory.
+
+Because it ships inside this package, the skill can never describe a command
+the installed CLI does not have. It is two files, `SKILL.md` and a command
+reference the agent opens only when it needs a flag, and they land in a
+`verifaied` folder inside the skills directory — yours to read and, in the
+project-local case, to commit for the rest of the team.
+
+Re-running the command writes nothing when the installed copy already matches.
+A copy that differs, because you edited it or because you have since upgraded
+the CLI, is left alone until you pass `--force`, which replaces the files
+verifAIed ships and never removes anything you added alongside them. Install
+somewhere else entirely with `--dir <path>`.
+
+Exit codes: `0` installed, updated, or already up to date; `1` usage error
+(`--user` with `--dir`, or a differing copy without `--force`).
+
 ## Agent loop
 
-We've found the most fruitful way to use verifAIed is to put a short
-loop into your coding agent's instructions file (`CLAUDE.md`,
-`AGENTS.md`, `.cursorrules`, `GEMINI.md`,
-`.github/copilot-instructions.md`) so it self-corrects on every change.
+Agents that do not read `.claude/skills/` want the same loop in their
+instructions file (`CLAUDE.md`, `AGENTS.md`, `.cursorrules`, `GEMINI.md`,
+`.github/copilot-instructions.md`), so they self-correct on every change.
 
 The easiest way is to let your agent write that loop into the rules
 file for you. Paste the prompt below into your agent — it will inspect
@@ -1000,7 +1046,7 @@ After implementing or modifying any code in this repo, run this loop until verif
    ```
 2. Upload the results to verifAIed:
    ```
-   verifaied upload
+   verifaied upload --junit junit.xml
    ```
 3. Ask if you're done: call the `check_done` tool on the verifAIed MCP server (or run `verifaied check-done`). It returns `{done, status, gaps}` for this branch — failing tests plus untested/partially-covered functions among the ones you changed.
 4. If `done` is true, stop — the branch is covered.

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "verifaied"
-version = "0.25.0.dev53"
+version = "0.28.0.dev71"
 description = "Find what's untested in your code — locally, no account required"
 readme = "README.md"
 requires-python = ">=3.10"

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/src/verifaied/cli.py

@@ -45,7 +45,20 @@ from verifaied.instrument import run_instrument
 from verifaied.prompts import build_prompt, format_line_ranges
 from verifaied.proxy import DEFAULT_PROXY_PORT
 from verifaied.record import DEFAULT_INTERVAL, DEFAULT_PORT, run_record
-from verifaied.repo_config import CONFIG_PATH, ENV_KEY, SECTION, RepoConfigError
+from verifaied.repo_config import (
+    CONFIG_PATH,
+    ENV_KEY,
+    SECTION,
+    RepoConfigError,
+    VideoProfile,
+)
+from verifaied.skillcmd import (
+    SKILL_NAME,
+    SkillInstallError,
+    detect_project_root,
+    install_skill,
+    resolve_skills_dir,
+)
 from verifaied.testcmd import (
     RecordedTestError,
     parse_test_ref,
@@ -85,6 +98,13 @@ var_app = typer.Typer(
 )
 app.add_typer(var_app, name="var")
 
+skill_app = typer.Typer(
+    add_completion=False,
+    no_args_is_help=True,
+    help="The agent skill that teaches a coding agent to drive verifAIed.",
+)
+app.add_typer(skill_app, name="skill")
+
 console = Console()
 err_console = Console(stderr=True)
 
@@ -1758,6 +1778,16 @@ def test_record_command(
         "--token",
         help=f"API token (vr_live_...). Defaults to ${ENV_API_TOKEN}.",
     ),
+    video: VideoProfile | None = typer.Option(
+        None,
+        "--video",
+        help=(
+            "How the replay that follows this recording should be paced. "
+            "`fast` marks each click and moves on; `watch` lingers on "
+            "every action; `off` records without annotations. Overrides "
+            "replay_show_actions_ms / replay_slow_mo_ms for this run."
+        ),
+    ),
 ) -> None:
     """Record one flow from the terminal, without a studio running.
 
@@ -1850,6 +1880,7 @@ def test_record_command(
             # No `--pre` means "leave the stored chain alone", which is a
             # different instruction from "compose with nothing".
             pre=list(pre) or None,
+            video=video.value if video is not None else None,
         )
     except (RecordedTestError, UploaderError) as e:
         err_console.print(f"[red]VerifAIed error[/red]: {e}")
@@ -1969,6 +2000,16 @@ def test_run_command(
         "--token",
         help=f"API token (vr_live_...). Defaults to ${ENV_API_TOKEN}.",
     ),
+    video: VideoProfile | None = typer.Option(
+        None,
+        "--video",
+        help=(
+            "How the video should be paced. `fast` (the default) marks "
+            "each click and moves on; `watch` lingers on every action; "
+            "`off` records without annotations. Overrides "
+            "replay_show_actions_ms / replay_slow_mo_ms for this run."
+        ),
+    ),
 ) -> None:
     """Replay a recorded test, capturing video, trace and coverage.
 
@@ -2024,6 +2065,7 @@ def test_run_command(
             proxy=proxy,
             proxy_target=proxy_target,
             proxy_port=proxy_port,
+            video=video.value if video is not None else None,
         )
     except (RecordedTestError, UploaderError) as e:
         err_console.print(f"[red]VerifAIed error[/red]: {e}")
@@ -2406,3 +2448,93 @@ def var_unset_command(
             "the environment variable it named is still exported wherever you "
             "set it.[/dim]"
         )
+
+
+def _display_path(path: Path) -> str:
+    """The shortest honest way to write a path: relative to cwd when it is under it."""
+    try:
+        return str(path.relative_to(Path.cwd()))
+    except ValueError:
+        return str(path)
+
+
+@skill_app.command("install")
+def skill_install_command(
+    user: bool = typer.Option(
+        False,
+        "--user",
+        help=(
+            "Install into ~/.claude/skills/ — every repository on this "
+            "machine — instead of this project's .claude/skills/."
+        ),
+    ),
+    directory: Path | None = typer.Option(
+        None,
+        "--dir",
+        "-d",
+        help=(
+            "The skills directory to install into, if it is neither this "
+            "project's nor your home one. The skill lands in a "
+            f"`{SKILL_NAME}` folder inside it."
+        ),
+    ),
+    force: bool = typer.Option(
+        False,
+        "--force",
+        "-f",
+        help="Overwrite an installed copy whose files differ from this one's.",
+    ),
+) -> None:
+    """Install the verifAIed skill, so your coding agent knows how to drive this.
+
+    The skill ships inside this CLI, so it always describes the commands
+    you actually have. It teaches an agent the loop — run the tests with
+    coverage, `verifaied upload`, `verifaied check-done`, fix what comes
+    back, repeat until done — and the rules that keep the loop honest.
+
+      verifaied skill install          # this project: .claude/skills/verifaied
+      verifaied skill install --user   # every project: ~/.claude/skills/verifaied
+
+    "This project" is the root of the repository you are standing in — run
+    it from a subdirectory and the skill still lands where your agent will
+    look for it.
+
+    Re-running it writes nothing when the installed copy already matches.
+    A copy that differs — because you edited it, or because you have since
+    upgraded the CLI — is left alone until you pass --force, which replaces
+    the files verifAIed ships and never removes anything you added
+    alongside them.
+
+    Exit codes:
+
+      0  installed, updated, or already up to date
+      1  usage error (--user with --dir, or a differing copy without --force)
+    """
+    try:
+        skills_dir = resolve_skills_dir(
+            user=user,
+            dest=directory,
+            cwd=detect_project_root(Path.cwd()),
+            home=Path.home(),
+        )
+        result = install_skill(skills_dir=skills_dir, force=force)
+    except SkillInstallError as e:
+        err_console.print(f"[red]VerifAIed error[/red]: {escape(str(e))}")
+        raise typer.Exit(code=1) from e
+
+    where = escape(_display_path(result.path))
+    if result.state == "unchanged":
+        console.print(
+            f"[green]Up to date[/green] — the verifAIed skill is already at "
+            f"[bold]{where}[/bold] ({len(result.unchanged)} file(s))."
+        )
+        return
+
+    verb = "Updated" if result.state == "updated" else "Installed"
+    console.print(f"[green]{verb}[/green] the verifAIed skill → [bold]{where}[/bold]")
+    for rel in result.written:
+        console.print(f"  {escape(rel)}")
+    console.print(
+        f"\n[dim]Your coding agent picks it up on its next session — ask it to "
+        f"use the {SKILL_NAME} skill.[/dim]"
+    )

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/src/verifaied/recorder_driver.js

@@ -48,9 +48,9 @@
 // and derives nothing from it: the spec goes out verbatim and Python
 // owns the single parser, so the two halves can never drift.
 //
-// Every Playwright API below is public and verified on 1.60, which is why
-// the Python side gates recording on a version **floor** rather than an
-// allowlist. Two startup probes still run — the console API and the AI
+// Every Playwright API below is public and available from 1.59, which is
+// why the Python side gates recording on a version **floor** rather than
+// an allowlist. Two startup probes still run — the console API and the AI
 // snapshot mode — because a browser that opened without either would
 // record nothing while looking like it worked.
 //
@@ -739,7 +739,7 @@ async function main() {
     );
   }
   // The assistant's whole view of the page is an AI-mode snapshot, and
-  // every element it names is a ref out of one. Public API in 1.60, but
+  // every element it names is a ref out of one. Public API from 1.59, but
   // probed because the failure mode of a build without it is a snapshot
   // with no refs — and an assistant that can see the page and never touch
   // anything on it.
@@ -1384,7 +1384,7 @@ async function main() {
    * snapshot measures two seconds of blocked main thread on a
    * 92k-node app — and the only thing that reads the tree is the
    * studio's assistant, so the CLI says when somebody is going to.
-   * `page.ariaSnapshot()` is public API in 1.60 and needs no probe; the
+   * `page.ariaSnapshot()` is public API from 1.59 and needs no probe; the
    * worst it can do is throw on a page that navigated mid-read, which
    * reports `page: null` and costs the next beat.
    */

{verifaied-0.25.0.dev53 → verifaied-0.28.0.dev71}/src/verifaied/recorder_refs.js

@@ -12,14 +12,14 @@
 // same one the recorder writes for a click. What the browser runs is what
 // the spec saves.
 //
-// How refs behave in 1.60, because the checks below rest on it: a ref is
-// cached ON THE ELEMENT and reused across snapshots while its role and
-// accessible name hold, so an element keeps its number as the page moves
-// around it — but the counter lives in the document's injected script and
-// starts again at e1 after a cross-document navigation, where the same
-// number can name a different element on a similar page. So a ref written
-// against an earlier snapshot is trusted when the document is the same and
-// its role and name still match, and refused otherwise.
+// How refs behave from 1.59 onward, because the checks below rest on it:
+// a ref is cached ON THE ELEMENT and reused across snapshots while its
+// role and accessible name hold, so an element keeps its number as the
+// page moves around it — but the counter lives in the document's injected
+// script and starts again at e1 after a cross-document navigation, where
+// the same number can name a different element on a similar page. So a
+// ref written against an earlier snapshot is trusted when the document is
+// the same and its role and name still match, and refused otherwise.
 //
 // Everything here is string work: finding the refs a line names, reading
 // what a ref stood for in a given snapshot, checking one line's identity