mind: date every active and issued task, and give the dashboard a Recent table

[Jammy2211]

The Mind dated only what it finished. Every complete/ record carries completed:, but a task that had merely been picked up carried no date a reader could parse — active.md said it in prose when a session remembered to (- issue: …/1501 (issued 2026-08-19)), which nothing could read back. So the Mind could answer "what shipped in July?" and not "what did we start?".

The convention

One machine-readable date field per registry entry, whose key names the event, plus the prompt's own copy in its light header:

Where	Field	The event it dates
active.md	- issued: YYYY-MM-DD	the day the task got its GitHub issue
planned.md	- filed: YYYY-MM-DD	the day it was scoped
parked.md	- parked: YYYY-MM-DD	the day it stopped
active/<name>.md	Issued: YYYY-MM-DD	the prompt's own copy, so an orphan stays dated

Reading is tolerant — registered:, started:, planned:, found: and shipped: all count, because the registries are hand-edited by many sessions and an entry that says when it happened should count however it said it. A date sitting in another field's prose deliberately does not: that is the un-parseable habit this replaces. Full schema in REFERENCE.md → "Task dates".

Retroactive backfill

scripts/lifecycle.py dates reports every entry and issued prompt carrying no date; dates --write reconstructs them from evidence the repo already holds, in order:

1. the commit that introduced the entry, or moved the prompt into active/
2. the dated registry entry that claims the prompt
3. a date the entry already stated in its own prose

Nothing is guessed — an entry with no evidence is reported for a human to date by hand. Every inferred date says where it came from, so a wrong backfill is auditable rather than anonymous:

Issued: 2026-08-18 (backfilled from parked.md `parked:`)

Shallow clones are not evidence. A shallow checkout (CI, a cloud session) reports its boundary commit as the day everything older appeared, so every backfill in one clone would come out the same wrong date. Git dates at or before the boundary are discarded and the chain falls through to the Mind's own records — which is exactly what happened writing this change: git offered 2026-08-21 for all four prompts, and the fallback produced 2026-08-08/18/19.

All four issued prompts are backfilled and active.md carries its canonical issued:; lifecycle.py dates now reports OK.

Where the dates get written from now on

/create_issue writes both the registry issued: and the prompt's Issued: header at the one moment anyone knows the date for certain, rather than leaving it to be reconstructed later.

The dashboard payoff

Regenerated dashboard.md / dashboard.html pick up the new Recent table (between Backlog and Epics) and the date facet now carried by In flight / Parked / Planned rows. The renderer itself is the paired PyAutoBrain change.

Testing

• tests/test_lifecycle_dates.py — 17 new tests: the key naming the event, the most-specific event winning, prose not counting, the backfill's sources and its refusal to guess, insertion that never splits a wrapped header value or a Repos: bullet list, and a real shallow-clone fixture proving the boundary is refused.
• python3 -m pytest tests/ — 176 passed.
• lifecycle.py check, lifecycle.py dates, repos_sync.py --check — all clean.

---

Generated by Claude Code