feat(stacking): move the stacking skill to pox-5

[biwasxyz]

Closes #429. Skills-side counterpart of aibtcdev/aibtc-mcp-server#682 (part of aibtc-mcp-server#677).

Why

pox-5 has been the active PoX contract on mainnet since cycle 141 and removed every function the stacking skill called. #424 made writes refuse; this adds real pox-5 staking.

What

Ported from aibtc-mcp-server#682, which I checked against mainnet before porting: its stake, stake-update and unstake arguments and post-conditions match real successful transactions, and its reward-claim post-conditions match the Xverse signer manager's payout code.

Subcommand	pox-5 call
get-pox-info	get-pox-info + burn height: cycle, next cycle start, prepare phase
get-stacking-status	get-staker-info, get-bond-membership, locked/unlocked balance
list-signers (new)	walks the signer set for a cycle; optional claim style per manager
stack-stx	stake(signer-manager, amount, num-cycles, start-burn-ht, signer-calldata)
extend-stacking	stake-update: extend, increase, switch signer, change payout calldata
unstake-stx (new)	unstake
get-rewards (new)	staker unclaimed + manager unpulled for a cycle
claim-rewards (new)	pulls for the manager if needed, then claim-staker-rewards

stack-stx / extend-stacking flags change: signer manager instead of a PoX BTC address and burn height. Optional --btc-reward-address (+ --max-withdrawal-fee-sats) encodes {pox-addr, max-fee} calldata; --signer-calldata-hex passes raw calldata.

Differences from the MCP version

• Every write first confirms /v2/pox still reports pox-5, and refuses before signing otherwise (fail closed if unreadable), keeping fix(stacking): refuse writes when pox-4 is not the active PoX contract; read pox-5 staking status #424's protection for a future PoX bump.
• StackingService takes an optional { hiro, callContract } for tests, instead of module mocks, which leak across files under bun test.

Also

• pillar-direct: direct-stack-stx --pool fast-pool and direct-revoke-fast-pool refuse while pox-4 is not active (the Pillar wallet contract hardcodes pox-4; the real fix is on Pillar's side). The Stacking DAO path is unchanged.
• @stacks/* → ^7.6.0 (staking / pox post-condition types). The lockfile change is limited to those packages.

Verification

• Mainnet reads: get-pox-info (cycle 143, prepare phase from 968350); list-signers → 26 signers for cycle 144 (same as the MCP PR); get-stacking-status for a real Xverse staker → 500 STX, unlock burn height 1170050, matching Hiro's burnchain_unlock_height; get-rewards → claim style staker-arg.
• src/lib/services/stacking.service.test.ts: the 16 MCP tests ported, plus 2 for the active-contract guard (verified to fail with the guard removed).
• CLI validation paths (amount 0, 97 cycles, both payout options, bad cycle) error before the wallet loads; Fast Pool commands refuse on mainnet before touching keys.
• bun run typecheck, bun run validate; stacking/stacking.ts and pillar/pillar-direct.ts type-checked separately (the repo's typecheck only covers src/).
• Full bun test (Bun 1.4.2): 180 pass / 38 fail vs main 162 / 38. The 38 failures are the existing cross-file mock leak; x402 files pass alone (72/72) on 7.6.0.
• Not exercised live: no transaction was broadcast. A small testnet stake → unstake before relying on it is worthwhile, as with the MCP PR.

🤖 Generated with Claude Code

[biwasxyz]

Review: rebased onto main (skills.json regenerated). Verified every pox-5 call against the live mainnet contract, its interface, 6 signer-manager contracts and ~450 real stake/stake-update/unstake txs: function names, arg order, calldata encoding (byte-identical to a real Xverse stake), pox-addr versions, post-conditions (Deny everywhere), cycle math and the fail-closed pox-5 guard all check out. Live mainnet reads match Hiro (status for a 12,000 STX staker: lock + unlock height 1172150 exact).

Fixed in 01665a6:

1. get-rewards / claim-rewards reported 0 and refused to claim until the manager pulled — get-earned-staker-rewards reads the signer's rewards-per-token, which only advances on the pull. Now projects via pox-5's compute-earned-rewards over the cycle's global rpt. Live: xverse-3 staker SP3WJ7…ZC7F cycle 144 now shows 336 sats (was 0).
2. extend-stacking silently deleted the BTC payout address — Xverse/Fast Pool validate-stake! map-delete the pox-addr when calldata is none (planbetter rejects it). Now requires --btc-reward-address / --signer-calldata-hex / explicit --sbtc-payout.
3. claim pull + claim could collide on nonce — consecutive nonces pinned.
4. Testnet claims always aborted — reward post-conditions now use /v2/pox pox_5_sbtc_contract.
5. 3-block margin before the prepare phase; error messages name CLI subcommands, not MCP tools.

+4 tests (22/22 stacking, 196 in src/lib). Known, not fixed: stake during an sBTC-bond rollover would abort on the uncovered refund (costs the fee, no loss); balance checks don't reserve the tx fee. No live broadcast was done — a small testnet stake → unstake is still worth running.