ADLC5 Pulse

Pulse is an on-demand snapshot of ADLC5 feature progress, recorded evidence, usage coverage, and a suggested next action. It uses the existing ADLC5 state and evidence and does not run consumer commands, change lifecycle state, or continuously monitor a repository.

./scripts/adlc5 pulse snapshot --workspace .
./scripts/adlc5 pulse band --workspace . --feature my-feature
./scripts/adlc5 pulse board --workspace . --open
./scripts/adlc5 pulse handoff --workspace . --feature my-feature

snapshot and band are read-only. Every board command writes the shared offline page at ~/.adlc5/pulse/board.html, with derived caches keyed by worktree under ~/.adlc5/pulse/workspaces/. The page includes discovered projects and all recorded host badges per worktree: Claude Code, Codex, Cursor and Hermes Agent. These reflect observed usage, not live presence. Rerun board to refresh it. handoff prints a summary for review; it does not write to a clipboard or send it elsewhere.

Navigate the board

Use Projects to select an application, Project to compare its worktrees, Workflow to inspect one feature, and paginated Activity for historical records. Settings shows effective machine, project, and worktree values, their provenance and overrides; it is read-only and excludes credential values. See configuration and storage.

Evidence is current at capture time. A local pr-ready gate observation is not remote CI, human approval, deployment, or production health. known_cost_usd is an attributable subtotal, not an invoice; complete_total_usd stays null because ledger coverage is unproven. Missing, unavailable, or unpriced records remain unknown. See the full Pulse guide for security, data, and plugin details.

The same skill and read-only snapshot tool ship in the Codex, Claude Code, Cursor, and Hermes Agent packages. Claude Code also displays a native interactive band above the prompt, refreshed at session start and after completed turns. Review and Audit invoke existing skills; Board writes and opens the offline board; Details expands the captured evidence. Codex and Cursor use the shared skill/MCP/CLI interfaces because their documented plugin surfaces do not expose a persistent custom prompt band. Claude may still use an optional cached status line configured manually. Live host qualification remains separate from package tests.

Savings and quality

The card and board show outcome success with sample counts, projected cost savings from observed comparable runs, escaped defects, and rework. Missing evidence displays Not measured. Outcome success is not a general accuracy score. The board shows a 95% Wilson interval for the recorded success proportion and an observed case range for savings; neither proves future performance.

Place actual evaluator exports (telemetry.jsonl, usage.jsonl, outcomes.jsonl) under .adlc5/<feature>/evaluation/. Savings require explicitly matched direct/candidate cases, equal repetitions, complete reported costs, and every run passing the existing quality floor. The experiment owner declares comparability with comparison_id; Pulse does not verify experimental design. No evaluation data is uploaded. See the full Pulse guide for the record contract.