AI Agent Workflow Collection

Log the current Claude Code session as a concise YYMMDD-HHMM- .txt under ~/Daily/YYYY-MM-DD/, then sync to the private jluo41/Daily git repo (pull --rebase → commit → push). Use when a working session ends OR on the user's explicit "log this session" request. The .txt is the input to chronicle-daily's evening rollup. Day-close aware — if today's folder has .closed, write to tomorrow instead.

Use when user requests diagrams, flowcharts, architecture charts, or visualizations.

Run any intervention-lifecycle work (the application umbrella). Use /haipipe-application enter or status to preload an open-needs dashboard from STATUS.md, 0-lifecycle, 0-artifacts, 1-rounds, and git state. Application lifecycle owns intervention-specific story, the stage-1 evidence ladder (1a-descriptions -> 1b-themes -> 1c-claims -> 1d-advice), narrative, displays, artifact text, maturity, and dated work rounds; the venue (sms/email/dashboard/report/...) gates which stages fire and how deep claims must settle; open evidence questions are RAISED as entries in the flat probe pool 1-probes/PPNN .md, and each stage's PROBE phase (haipipe-application-probe) binds them to answers in the task/discovery bank through a clean collector agent — never calling the bank directly. Trigger: application, intervention, enter, status, seed, ladder, descriptions, themes, claims, advice, venue, pitch, narrative, display, section-edit, draft, sms, message, email, dashboard, report, review, deploy, iterate, round, probe, /haipipe-application.

Coverage / readability scanner of the haipipe-insight family. Reads the project's probes/ and existing insights/ folders; reports which probes are CONFIRMED and ready for synthesis, what's already in the insight base, and which gaps a session could close. NO code execution. Use to plan the next /haipipe-application ask, or as a standalone read-only audit. Trigger: explore, scan, coverage, what can we synthesize, what's missing, /haipipe-insight-explore.

Use when writing or revising ML or AI conference papers for venues such as NeurIPS, ICML, ICLR, ACL, AAAI, or COLM, especially when the workflow is conference-first rather than Nature-style journal-first.

QA specialist of haipipe-probe. Three complementary checks. (1) STRUCTURAL: audits run quality (per-run sanity) and probe quality (statistical claim integrity) via checklists, produces ✅/⚠️/❌ + actionable issues. (2) INTEGRITY: Codex MCP fraud-pattern audit (ground-truth provenance, metric-definition consistency, phantom results, scope-language mismatch, individual/split leakage), writes INTEGRITYAUDIT.md. (3) SEMANTIC: Codex MCP judges whether evidence supports the intended claim (yes/partial/no + confidence), writes CLAIMSFROMRESULTS.md. The honest-science gate before a claim becomes a paper-able statement. Trigger: review, audit, qa, integrity, fraud check, fake ground truth, phantom results, scope check, claim verdict, supports?, /haipipe-probe-review.

Task-folder and task-group orchestrator. For a task-folder: runs the 4-stage code lifecycle (Plan → Build → Execute → Report) or dispatches to type specialists for scaffolding. For a task-group: iterates over each child task-folder and runs the lifecycle on each one. For insight (filing Ddata observation cards from results), use /haipipe-insight with the task-folder path.

agent task-folder build specialist. Scaffolds {NN} / task-folders under F-series task-groups that call an LLM agent with prompts + tools — outputs to results/ /{transcript.json, summary.md}. Called by /haipipe-task orchestrator when task-type=agent. No corresponding pipeline skill yet.

individual-query task-folder build specialist. Scaffolds {NN} / task-folders under E-series task-groups that query / visualize ONE individual's data (CGM trace, meal timeline, treatment events) — outputs to results/ /{plot.pdf, table.csv}. Called by /haipipe-task orchestrator when task-type=individual. Cross-references /haipipe-individual.

Playbook for the Related Work section. Positioning angles (axis-of-comparison / chronological / closest-prior), how to avoid lit-review-creep, when to use a comparison table. Use when writing or revising 0-sections/02relatedwork.tex (or equivalent). Trigger: related work, lit review, positioning, /section-related-work.