Exploration
These three skills navigate the real application (with the Playwright MCP browser tools; the exploratory session uses whichever interaction capability fits the target, typically the browser) to validate a plan, confirm whether a bug reproduces, or run a dynamic exploration session guided by a charter — always documenting what they find with evidence (screenshots, steps, expected vs. actual result).
lippia-neo-run-manual-testing
When to use it
To manually explore a web application with the Playwright MCP browser tools and
capture structured findings: run a plan's scenarios, check expected
vs. actual result, document UI inconsistencies, missing validations, and unexpected
behaviors, with screenshots as evidence. It's agnostic of the project's testing framework
(it always uses Playwright MCP) and it's optional in the flow: lippia-neo-generate-test-suite
can run without its findings, although the quality of the generated tests improves if it has them.
{
"input": {
"test_plan_path": "test-plans/PA-US-02.md"
}
}
| Input | Type | Description |
|---|---|---|
input.test_plan_path |
Required | Path to the test plan to run (it can also receive inline scenarios via input.scenarios, or go straight in via input.free_exploration). |
input.free_exploration |
Optional | Free-form description of an ad-hoc scenario to explore (a bug, a standalone flow), with the suggested shape { description, expected, actual, url }; if present, the skill synthesizes a minimal scenario and skips input.test_plan_path. |
input.us_id |
Optional | Identifier that defines the name of the output folder; effectively mandatory when going in via input.free_exploration (there's no US-ID to derive in that case). |
Output
| Field | Description |
|---|---|
test_plan_path, us_id |
The US-ID actually used. |
report_path |
The human-readable .md. |
manual_execution_report_path |
The handoff .json, which the caller must consume as-is. |
executed_scenarios |
Per scenario: id, title, status (pass/fail/blocked/skip), and step-by-step result. |
findings |
Each full anomaly: id (ANOMALY-NNN), type, classification, severity, priority, screenshot, context, environment, URL, reproduction steps, expected/actual. |
ui_selectors |
Real selectors validated during exploration. |
screenshots |
List of screenshots. |
summary |
Totals of scenarios and findings (including critical ones). |
Chains with
Comes after lippia-neo-create-test-plan; its output (manual_execution_report_path, ui_selectors) feeds
lippia-neo-generate-test-suite.
lippia-neo-reproduce-bug
When to use it
When you need to confirm whether a bug reproduces — it navigates the live app with Playwright MCP
following only the steps and expected behavior described in the source (a
user story, or a bug report with or without a linked story) and reports whether the bug reproduces
or not. It doesn't diagnose the root cause or propose a fix. It typically runs as step 3 of a
bug cycle, right after lippia-neo-find-existing-tests. It requires Playwright MCP to be configured (it aborts if
the browser_* tools aren't available) and it forbids mutating the DOM by hand or verifying a state
in the same call that caused it.
{
"input": {
"source_summary": {
"bug_id": "BUG-104",
"source_kind": "jira"
}
}
}
| Input | Type | Description |
|---|---|---|
input.source_summary |
Required | Output of lippia-neo-read-user-story or of lippia-neo-read-bug-report; determines whether the "story", "bug without a story", or "bug with a linked story" case applies. |
input.issue_body |
Optional | Raw text of the bug/issue, only to identify the acceptance criterion at stake and the steps the reporter tried — never to make up steps. |
input.related_us_summary |
Optional | If source_summary.related_user_story is non-null and has already been resolved with lippia-neo-read-user-story; it gives access to that story's real AC/key_features. |
input.max_attempts |
Optional | Cap on retries with explicit variants from the story; defaults to 3. |
Output
| Field | Description |
|---|---|
bug_reproduced |
Boolean. |
acceptance_criterion_violated, reproduction_steps, expected_result, actual_result |
Literal, without reinterpretation. |
evidence |
screenshots, console_errors, network_errors. |
environment |
url, browser, app_version. |
attempts |
Each attempt with attempt_number, variant, result, notes. |
not_reproduced_reason |
Mandatory if bug_reproduced: false. |
report_path, scripted_execution_handoff_path (only if it reproduced), ui_selectors, summary |
Chains with
Follows lippia-neo-read-user-story/lippia-neo-read-bug-report; if bug_reproduced: true, it writes the
handoff that lippia-neo-generate-test-suite consumes, followed by lippia-neo-run-and-heal and
lippia-neo-write-test-report. It's the skill that replaces lippia-neo-run-manual-testing when the specific
goal is to confirm a bug.
lippia-neo-run-exploratory-session
When to use it
When there's exactly one bounded, approved exploratory charter (CHARTER-<NNN>) ready to
run, or when a previous session needs to be resumed from its checkpoint. It runs a live,
time-boxed SBTM session against the real application, recording findings, evidence,
opportunities, checkpoints, and a resumable report. It designs and runs the tests
dynamically based on the mission (the charter's focus isn't a checklist), and an unexpected
behavior isn't classified as a defect without a valid basis. It runs a single charter — it never builds or
orchestrates a full portfolio (that's done by lippia-neo-meta-run-exploratory-testing).
{
"input": {
"artifact_scope_id": "EXP-20260101-100000",
"charter_path": "exploration/EXP-20260101-100000/charters/CHARTER-003.md"
},
"config": {
"paths": {
"exploratory_dir": "exploration/"
}
}
}
| Input | Type | Description |
|---|---|---|
input.artifact_scope_id |
Required | Functional scope to preserve (never the engine's technical run_id). |
config.paths.exploratory_dir |
Required | Root directory for exploratory artifacts. |
input.charter / input.charter_path |
Optional | The charter to run (one of the two); mandatory for a new session. |
input.resume_report_path |
Optional | To resume an interrupted session from its checkpoint/report (it reuses the same SESSION-<NNN>). |
input.report_language |
Optional | Language of the visible Markdown (es/en). |
input.allow_persistent_mutations |
Optional | Defaults to false; enables actions that create/delete/confirm persistent data. |
config.environment / config.auth / config.paths.env_file |
Optional | To resolve credentials and environment context. |
Output
| Field | Description |
|---|---|
artifact_scope_id, charter_id, session_id, session_status (completed/in_progress/blocked), resume_required |
|
blocking_reason |
none/interaction_capability_unavailable/target_unavailable/auth_unavailable/invalid_charter/unsafe_mutation_required/runtime_interrupted/unrecoverable_error. |
elapsed_session_minutes, target_time_box_minutes, report_language, report_path, report_json_path, checkpoint_path |
|
session_log (T/B/S category per entry), time_distribution (test_pct/bug_pct/setup_pct) |
|
findings |
Each one with id, classification (defect_candidate/observation/false_positive/change_request/pending), title, description, expected_basis, expected_result, observed_result, severity and priority (they can stay unclassified), discovery_context, evidence. |
opportunities_discovered |
Each one with observation and suggested_charter. |
open_questions, evidence |
|
coverage_summary |
Qualitative summary of what was explored, what was left uncovered, and whether a follow-up charter is worthwhile (no high/medium/low label). |
Chains with
Consumes a charter produced by lippia-neo-create-exploratory-charters; it's step 3 of
lippia-neo-meta-run-exploratory-testing, invoked once per charter (with resume
re-invocations if it stays in_progress).