go <dir> refuses a plan that already ran, and the record stops forgetting

[Shashankss1205]

Closes #100.

The bug

Bare grapharc go has always skipped executed plans — find_unexecuted_plan passes over any record carrying an executed_run_id. The explicitly-named-directory form made no such check:

grapharc plan "look into the outage" --scripted --trace r5/p.jsonl
grapharc go r5   # exit 0
grapharc go r5   # exit 0, runs the whole graph again
grapharc go r5   # exit 0, and again

plan.json executed_run_id : f81e7318e9dc
runs in the trace         : ['b563daf1c7db', '1f792472a215', 'f81e7318e9dc']
3 executions happened; plan.json names 1

Two fixes, and the second is the load-bearing one

The record accumulates. executed_run_ids (oldest first) plus an executed_at stamp. The scalar stays as the newest — find_unexecuted_plan and grapharc/mcp/driver.py read it, and a plan.json written before this change must keep working. _executed_run_ids() falls back to the scalar, so an old record reports its one run rather than reporting none; a malformed list degrades to the same fallback rather than raising, because this is a record for a human to read.

One approval can no longer be spent N times. An approval binds to a proposal fingerprint, which does not change between runs of the same saved plan. Re-issuing go <dir> on a mutating: true plan was an agent editing the tree once per invocation on the strength of a single human yes. A plan carrying an executed_run_id is now refused with exit 2 — before anything executes, naming the previous run and its timestamp — unless --again asks for the re-run explicitly.

Bare grapharc go behaves exactly as before: it skips executed plans quietly rather than refusing a directory it was never given.

The one test that had to change

test_go_and_plan_share_every_planning_flag asserts go_actions - plan_actions == set(). --again is go-only by design — it governs re-executing a saved plan, and plan never executes one, so it is not a planning flag. The exemption is recorded in the test with its reason, mirroring how --scripted / --go are already recorded as plan-only.

Verification

• uv run pytest — full suite green (exit 0).
• uv run ruff check grapharc tests — clean.
• The new tests can go red: neutering the guard and the accumulation while leaving the helper importable turns 4 of the 8 red (test_a_second_go_on_an_executed_plan_is_refused, test_the_refusal_names_the_run_and_when_it_happened, test_the_record_names_every_run_that_executed_the_plan, test_an_old_record_is_refused_and_then_accumulates_from_its_scalar).

Out of scope, as the issue asks: the fingerprint protocol, and whether a re-run of a mutating plan should re-park even without --approve.

🤖 Generated with Claude Code