Durable PM/TPM charters get a real wake cycle and CLI surface

The Broader Goal at the End of the Previous PR

The pm and tpm roles existed as charter files but had no durable wake-cycle identity — load_charter()'s missing-file fallback hardcoded harness_bindings for tpm/architect to ["claude", "codex"], and neither role had an agent run path, a cron-installable entry, or CLI subcommands of their own. The intake→triage→approval loop (raw GitHub issues → TPM triage → PM approval of the drafted queue) existed conceptually but had no code path a scheduler could actually invoke.

Strategic Shifts in This PR (if any)

None — this PR is the direct, planned continuation of docs/superpowers/plans/2026-10-13-finalize-workspace-agent-charters.md (story-0ad6c02b, GOVERNS goal-a222b393/goal-06758149). The one adjustment made mid-plan: after two consecutive dispatched jobs (Tasks 7 and 8) silently no-op'd while reporting plausible success and cost, the remaining fully-deterministic, zero-judgment tasks were implemented directly rather than risking a third wasted dispatch cycle — documented in each commit and filed as gh:#2245.

What This PR Shipped

Eleven tasks, executed via Subagent-Driven Development (implementers dispatched to Codex/Claude per the project's role split, reviewer/verification stages retained):

• Charters (Tasks 1-4): synlynk/charters.py gained durability/wake_cycle_sop fields and now tolerates missing harness_bindings. .synlynk/charters/pm.json and .synlynk/charters/tpm.json were written as durable charters with 7-step wake_cycle_sop specs — autonomous authorities (triage_issue/draft_plan/dispatch_track for TPM) and explicit escalation triggers (technical_ambiguity/priority_ambiguity/dispatch_blocked_twice). maintainer.json was promoted to the same durability tier.
• Wake-cycle logic (Tasks 5-6): synlynk/tpm_sweep.py::run_tpm_wake_cycle() and synlynk/pm_agent.py::cmd_pm_wake() implement the actual intake/triage/approval loop against backlog_items, with new PM-approval-gate columns and synlynk/db_schema.py migration support.
• Agent configs and routing (Tasks 7-8): .agents/pm.json / .agents/tpm.json let _load_agent_config() resolve the two roles; synlynk/support_engineer.py::cmd_agent_run() now special-cases name == "pm"/"tpm" to call the wake-cycle functions directly, ahead of the generic signal-collector path.
• CLI surface (Task 9): synlynk pm wake and synlynk tpm wake subcommands, wired into both the argparse subparser registration and the dispatch elif chain in synlynk/cli.py. Fixing this surfaced a real regression — COMMAND_TAXONOMY had no entries for either new leaf command — closed by adding two entries in synlynk/taxonomy.py alongside the existing pm sweep/tpm sweep rows.
• Verification-only (Tasks 10-11): confirmed _install_cron_entry() is name-agnostic and needs no pm/tpm-specific branching (it already generically shells out to agent run {name}, which now hits the new wake-cycle branch); ran the full test suite (4109 passed, 52 failed/18 errors — all independently confirmed pre-existing sparse-checkout/environment artifacts, none caused by this branch) and confirmed .synlynk/policy.json and the qa/designer/marketing agent configs (which don't exist) are untouched vs origin/main.

A new ticket, gh:#2245, documents a suspected dispatch/logs diff-reporting bug: two dispatched jobs reported real cost and a plausible "files touched" list while making zero actual changes, and both reported lists were byte-for-byte identical to each other despite very different task text — pointing at a stale diff-base in the reporting path rather than a harness failure.

Brainstorm Visuals Used

None — this PR executes an already-approved implementation plan; no new design exploration was needed.

What This Achieved on the Path to Autonomy

The PM/TPM roles are no longer charter-only placeholders. A scheduler (cron, or a future daemon wake trigger) can now invoke synlynk agent run pm / synlynk agent run tpm — or a human/agent can call synlynk pm wake / synlynk tpm wake directly — and get the actual triage/approval loop, not a signal-collector pass that doesn't know what "pm" or "tpm" durability means.

Strategic Note: The Goal at the End of This PR

The wake-cycle code paths exist and are tested, but have not yet run against a live GitHub issue backlog outside of unit-test fixtures — that's the natural next step once this PR is reviewed and merged. gh:#2245 (dispatch diff-reporting bug) is filed but not yet investigated; it's worth resolving before relying on dispatch "files touched" output for no-op detection in future plan executions.

Related Docs

• docs/superpowers/plans/2026-10-13-finalize-workspace-agent-charters.md
• GOVERNS story-0ad6c02b, goal-a222b393, goal-06758149
• gh:#2245