Affected Repositories

• PyAutoCortex (primary)
• PyAutoBrain

Work Classification: Library
Worktree root: ~/Code/PyAutoLabs-wt/cortex-tasks-not-phases/

Branch Survey

Unregistered overlap: worktree cortex-may-submit holds open PRs PyAutoCortex#17 and PyAutoBrain#360 (docs-only, same files). Decision 2026-09-07: fold their wording into this branch; close both unmerged.

Suggested branch: feature/cortex-tasks-not-phases

Implementation Steps

1. PyAutoCortex script + tree. git mv phases/ tasks/; in scripts/cortex.py rename PHASE_FILE_RE, load_phases, phase_problems, Phase class and every user-facing message; delete the Phase: header and the --phase flag on new; add required Summary: (≤10 words) to the template and to check; ruling head field Phase: → Task:; ruling title # R-… — <verb> <project> <slug>; "revival is a new slug". Update every task file header (drop Phase:, add Summary:, title # <Project> — <title>) and all 29 rulings' Task: paths. Docs: AGENTS.md (fold in docs: a submission is made on the human's ask, by human or agent #17), REFERENCE.md, README.md, docs/schema_decisions.md, tests/.
2. PyAutoBrain conductor. agents/conductors/cortex/_cortex.py: census keys phases → tasks; SECTIONS loses live, ready, gated; stats tiles and Summary table lose Ready (and the "Next phase" column becomes "Next task"); by-project block renders every open task as its own _html_task row headed by Summary: with a state pill (tone map planned/ready/accepted g, gated n, submitted/running/pulled/awaiting-ruling y, any failed/timeout/void run r) and only that state's facets; delete _fold_label; paths as <span class="pathchip"> with solid var(--accent) background and var(--accent-ink) text in _FRESH_CSS. Fold in #360's wording across AGENTS.md, AUTONOMY.md, agents/conductors/cortex/{AGENTS.md,cortex.sh}, skills/COMMANDS.md, skills/cortex/{SKILL.md,cortex.md}. Intake pill "gates a Cortex phase" → "gates a Cortex task". tests/test_cortex_conductor.py follows.
3. Summaries (prose). One Summary: per task file (~40): the question the task answers, ≤10 words, no project name, no number.
4. Acceptance. python3 scripts/cortex.py check clean; pytest green in both repos; pyauto-brain cortex dashboard --apply twice, --check stable; grep -ri phase in PyAutoCortex returns only Mind-epic references and historical ruling prose.

Key Files

• PyAutoCortex/scripts/cortex.py — registry parser, check, move/rule/new/gates/retire
• PyAutoCortex/tasks/<project>/<slug>.md (was phases/) — task files
• PyAutoCortex/rulings/**/*.md — Task: head field
• PyAutoCortex/{AGENTS,REFERENCE,README}.md, docs/schema_decisions.md
• PyAutoBrain/agents/conductors/cortex/_cortex.py — census + both renderers
• PyAutoBrain/board/_theme.py — pill tones and --accent-ink (read only)
• PyAutoBrain/tests/test_cortex_conductor.py
• PyAutoBrain/skills/cortex/, agents/conductors/intake/_intake.py (gate pill wording)

Cortex: tasks not phases — one-line summaries, state colours, fewer sections

Type: feature
Target: pyautocortex
Repos:

• PyAutoCortex
• PyAutoBrain
  Themes:
• dashboard
• science-workflow
  Difficulty: medium
  Autonomy: supervised
  Priority: high
  Status: formalised
  Consequence: judge
  Review-minutes: 30
  Unattended: no
  Filed: 2026-09-07

Make the PyAutoCortex board read like a science project: a project holds
unordered tasks (ideas the human runs in whatever order results dictate),
each with a ten-word summary and a coloured state, all visible at once.

Original request (verbatim)

More feedback on PyAutoCortex making it suited to doing science. Firstly, I
think we should do away with the term "phases", which implies that tasks are
sequential, noting that phaes have numerical numbers. I don't think most
science analysis goes like that. Rather, it is generally the case that there
are tasks, representing ideas I have for the project, and I run them, but the
order may change and adapt based on how results come in or other priorities.
Secondly, Rathert hen "4 more open phase(s) · plans and issues" being a drop
down menu, I think that all active tasks for each project should be listed
and shown on the dashboard by default, however I think it is important that
the task takes up one line each and is more readable. Thus, for each task, I
think rather than it being listed like "Inference_programme — phase 11:
cluster extended-source inference — gradient-based fitting building on JAX
knowledge — planned" it should have a summary that is up to 10 words so I can
read is quickly. For conciseness, we dont need the "Inference_programme -
phase 11:" bit -- its under the inference_programme heasder. So for this one
I would do somehtig like "Cluster extended source inference testing gradient
based fitting". So, every task has a summary of up to 10 words, which makes
the dasboard a lot more readable. Thirdly, more color to indicate state is
good, for example "planned" could be in green if irts planned and ungdated.
"active" could be yellow if its going on, green when complete. Use red when a
run is broken and needs attention. For these, can we make the path font in
white, so it stands out against pink: Local
/home/jammy/Code/PyAutoLabs/autolens_profiling Mirror
/mnt/c/Users/Jammy/Science/inference_programme RAL
/mnt/ral/jnightin/autolens_profiling . Fourthly, think about this and tell me
if you think I'm wrong, but I think we can remove the "Ready" and "Gated"
sections, this imformation should come from the projects bit above which we
are now seeking to improve and make more visible. These sections just mean
project updates are spread over more places.

Decisions taken with the human on 2026-09-07 (all four points agreed; "yes
full rename lets do it properly", "yes remove as recommended"):

1. Full rename: phase → task, and the number goes

In @PyAutoCortex the identity of a task is its slug (the file name), which is
already unique per project — the number carries nothing.

• phases/<project>/<slug>.md → tasks/<project>/<slug>.md (git mv, every
  file). scripts/cortex.py PHASE_FILE_RE, load_phases, phase_problems,
  new, move, rule, gates, retire and their messages follow.
• Delete the Phase: N header outright (no deprecated shadow field). Title
  line becomes # <Project> — <title>; ruling titles # R-… — <verb> <project> <slug>; "revival is a new phase number" becomes "revival is a new slug".
• Rulings: Phase: head field → Task: pointing at the tasks/… path, in all
  29 rulings and in ruling_problems. Supersedes: logic unchanged.
• cortex.py new <project> <slug> loses --phase; the template writes
  Summary: (see 2) instead.
• Docs follow: AGENTS.md, REFERENCE.md, README.md,
  docs/schema_decisions.md, checkin.yaml if it names phases; tests/.
• Mind side stays: Mind epics really are sequential; Epic: on a task file is
  unchanged. Mind intake's "gates a Cortex phase" pill wording → "gates a Cortex
  task".
• In @PyAutoBrain: agents/conductors/cortex/_cortex.py (218 mentions), its
  AGENTS.md, skills/cortex/, tests/test_cortex_conductor.py, and the
  intake conductor's Cortex-gate pill. Census keys (c["phases"]) rename to
  tasks.

2. Summary: header, one line per task, no fold

• New required header Summary: — at most ten words, the question the task
  answers, not its method. cortex.py check enforces presence and the word
  cap. Write one for every existing task file (~40); this is prose, Opus tier.
• Under ## Projects every open task of a project is its own tap-to-copy row:
  Summary as the link, a state pill, then only the state's facts (run ids,
  wall vs budget, gate refs, ⚠️ failed runs). No project name, no number, no
  "N more open phase(s)" fold. Chips per state as today.

3. State colours and the path block

Pill tone map (shared board/_theme.py tones, no new CSS):

Paths (Local / Mirror / RAL): render as a solid accent chip with
--accent-ink (white) text instead of the pink-on-tint inline code, so they
read on both themes. Markdown twin unchanged.

4. Sections removed

Drop Ready, Gated and Running / submitted sections, their stats
tiles and the Summary table's Ready column; the Projects block now carries
that information per task with colour. Awaiting ruling stays as the one
cross-project list (a ruling is the human's act). Recent rulings stays.

Acceptance

• python3 scripts/cortex.py check clean; PyAutoCortex + PyAutoBrain tests
  green; pyauto-brain cortex dashboard --check stable across two renders.
• grep -ri phase in PyAutoCortex returns only Mind-epic references and
  history (ruling bodies may keep prose written at the time).
• The rendered board shows every open task per project, one line, coloured.