PST Run

End-to-end pipeline runner for PST. Chains all four stages in sequence using isolated subagents:

pst-explore → pst-plan → pst-generate → pst-debug (if needed)

When to use

When you have test case documentation and want to run the full pipeline from exploration to a passing test in one invocation.

Inputs

The user pastes test case documentation inline in the chat.

Steps

1. Collect inputs

Ask the user:

• Session name — short slug for this test case (e.g. sign-in)

Confirm the test case documentation is present in the conversation.

2. Run pst-explore

Invoke the pst-explore skill. Provide:

• The session name
• The full test case documentation text

Verify that .pst/sessions/[name]/insights.md was created before proceeding.

3. Run pst-plan

Invoke the pst-plan skill. Provide:

• The session name

Important: This stage requires a user interaction — the skill will ask the user to confirm or specify the test file path. Wait for the user's response before continuing.

Verify that .pst/sessions/[name]/plan.md was created before proceeding.

4. Run pst-generate

Invoke the pst-generate skill. Provide:

• The session name

If tests pass: Proceed to step 6.

If tests fail: Collect the full error output and proceed to step 5.

5. Run pst-debug — only if tests failed

Invoke the pst-debug skill. Provide:

• Session name
• Failing test file path
• Full error output from step 4
• Instruct it to use Mode 1 (from session — full context available)

Confirm whether the test now passes before reporting.

6. Report final status

PST Run complete.

Session:    .pst/sessions/[name]/
Test file:  [confirmed path]
Status:     PASSING | FAILING

Artifacts:
  insights.md  ✓
  plan.md      ✓
  snapshots/   ✓ ([N] files)

If still failing after debug, report:

PST Run ended with a failing test.
See pst-debug output above for diagnosis and next steps.

Rules

• Each stage runs in isolation — never combine stages
• Never skip a stage even if artifacts appear to already exist — always re-run to ensure freshness
• The pst-plan stage always requires user confirmation of the test file path — do not skip this interaction
• If any stage fails unexpectedly (not a test failure but an execution error), report the error and stop — do not auto-retry