dissenting:
Maples changed the law. Stokley asks us not for habeas relief, but to stay the mandate in light of this change and remand for full consideration of whether he can overcome procedural default on his colorable Eddings and Skipper claims that were not raised because Harriette Levitt abandoned him. The only analysis we should do here is to determine whether he has made a prima facie case for abandonment under Maples to establish cause, and shown that his prejudice argument has some merit in that he does not raise a frivolous claim. His claim that the Arizona Supreme Court committed causal nexus error in declining to consider mitigating evidence is anything but frivolous. It is a constitutional claim and one that this court should not extend itself to decide on the merits before it was briefed or argued by either party.
The majority assumes without deciding that there was a Maples error. Respectfully, that was the only question before this court. The majority brushes it aside to get to the final end game, but further confuses our law on prejudice and standards for error review in the process. Because I cannot agree with the majority’s approach, I strongly dissent.
I first address why Maples error exists in this case. Then I turn to the majority’s incorrect and unrestrained analysis of prejudice.
I. Stokley has shown abandonment
Maples is not limited solely to actual abandonment. To obtain the remand he *406requests, Stokley need only make a prima facie showing of abandonment under Maples that might constitute cause to overcome procedural default. See Moornann v. Schriro, 672 F.3d 644, 647-48 (9th Cir.2012). Despite the extremely limited briefing on the pending motion, Stokley has made such a prima facie case of abandonment. Moreover, as the majority recognizes, he has a colorable underlying constitutional claim. Our inquiry should end there. I would grant the motion and remand to the district court for determination of cause and prejudice and, if appropriate, the merits of Stokley’s constitutional claim.1
Maples rests squarely on agency principles. 132 S.Ct. at 922-24. To explain how an agency relationship may be actually or constructively severed, the Supreme Court relied on Justice Alito’s concurrence in Holland v. Florida, 560 U.S. -, 130 S.Ct. 2549, 177 L.Ed.2d 130 (2010), to distinguish attorney negligence from abandonment. “Common sense dictates that a litigant cannot be held constructively responsible for the conduct of an attorney who is not operating as his agent in any meaningful sense of that word.” 132 S.Ct. at 923 (citing Holland, 130 S.Ct. at 2568 (Alito, J., concurring)). Justice Alito’s concurrence in Holland also noted that the agency relationship is constructively severed “particularly so if the litigant’s reasonable efforts to terminate the attorney’s representation have been thwarted by forces wholly beyond the petitioner’s control.” Holland, 130 S.Ct. at 2568. Indeed, our court’s precedent—while not finding abandonment—recognizes that Maples rests on agency principles and that a serious breach of loyalty can sever the attorney-client relationship in a manner that may constitute constructive abandonment sufficient to establish cause. See Towery v. Ryan, 673 F.3d 933, 942-43 (9th Cir.2012) cert. denied, -—• U.S.-, 132 S.Ct. 1738, 182 L.Ed.2d 271 (2012) (separately analyzing two prongs of actual abandonment or “serious breach of loyalty” and distinguishing Holland, which involved violations of fundamental canons of professional responsibility, from Tower/s circumstances, which did not).
In light of Maples, it is now recognizable that Stokley’s situation in postconviction proceedings was worse than simply “unenviable.” 659 F.3d at 810. Here, the attorney-client relationship was irrevocably broken. Further, the record demonstrates that, once the state was successful in forcing it to be put back together, post-conviction counsel Harriette Levitt actively undermined the work of Stokley’s replacement counsel and prevented Stokley from investigating and raising his own claims. While it has no legal bearing on the present issue, I note at the outset that Harriette Levitt is the same attorney whose conduct was at issue in the Supreme Court’s recently-created ineffective assistance of counsel exception to the once settled rule in Coleman. Martinez v. Ryan, — U.S.-, 132 S.Ct. 1309, 182 L.Ed.2d 272 (2012). Whereas the petitioner in Maples “in reality ... had been reduced to pro se status,” 132 S.Ct. at 927, Levitt’s actions regarding Stokley’s attempts to fairly present his claims arguably left him in a situation worse than a pro se petitioner. If there were ever a case for constructive abandonment under Maples, this is it.
Levitt filed her first post-conviction petition eight months after being assigned to the case. During these eight months, she *407initiated no contact with Stokley. The only communication she had with Stokley was a twenty-minute collect phone call he placed to her. Levitt did not conduct any independent investigation during this period, other than a few telephone calls lasting less than a total of two hours. According to Stokley, Levitt did not even receive the trial transcripts until more than six months after her appointment, and after the deadline for filing Stokley’s petition had passed.
When Levitt finally filed Stokley’s petition, she raised only two claims and wrote only three and a half pages of legal argument. Levitt’s billing records indicate that, aside from reviewing Stokley’s file and transcript, she spent no more than ten hours researching and writing his petition for post-conviction relief. Stokley immediately recognized the inadequacy of the petition and called Levitt to object. Levitt told him that his “trial attorneys didn’t make any mistakes” and that he would “probably be executed in 2 or 3 years.”
Stokley then took every action he could think of to object to Levitt’s continued representation. He wrote a letter to the Superior Court judge, expressing his concerns about the brevity of the petition and Levitt’s lack of interest and diligence. He wrote that he found it “evident that my present appeal has been handled with a lick and a promise, rather than being given the conscientious analysis and preparation which should be applied.” He asked the court to “appoint an attorney who will apply his or her self and try to do a competent job in this matter.” He sent a similar letter to the Arizona Capital Representation Project asking for help. The Superior Court forwarded Stokley’s letter to Levitt but took no other action.
Stokley also filed a complaint with the State Bar of Arizona protesting Levitt’s handling of his case. The Bar overlooked the posture of Stokley’s case and responded that his complaint could be dealt with in post-trial proceedings, noting that “[i]f there [was] a judicial determination that the lawyer acted improperly, [the Bar] would review the matter at that time.”
Not surprisingly, the Superior Court denied Levitt’s two-claim petition. Levitt then filed a motion to withdraw as Stok-ley’s counsel, citing the Bar complaint filed against her. She wrote that “[t]here has ... been a complete breakdown of the attorney-client relationship.” The court granted the request and appointed Carla Ryan as replacement counsel.
The state immediately moved to reinstate Levitt as Stokley’s counsel. The state argued that the initial petition had already been denied, and so there was “no valid reason for ... paying yet another defense attorney to review the voluminous record for the first time.” The state argued in the alternative for the court to limit the scope of Ryan’s representation, arguing that, if replacement counsel were appointed, she should be forbidden to “supplement the already-adjudicated petition in some manner,” because Arizona rules “do not allow for any such thing.” Notably, however, the Arizona Supreme Court eventually did permit Levitt to file a supplemental Rule 32 petition, specifically allowing her to “raise any issue ... even though it may not have been included in her first petition for post-conviction relief.” The state also objected to Ryan’s request for co-counsel in an unprofessionally worded opposition, arguing that Ryan was requesting a “side-kick” to “milk[ ] this ease for all it is worth as a cash cow.... Capital litigation is not an unlimited pot-boiler for the enrichment of private attorneys.” The Superior Court ordered Levitt reinstated, over the objections of both Levitt and Stokley.
*408Ryan was Stokley’s attorney for only one month. During that month, she spent much of her time responding to the state’s attempt to have her removed as counsel. Ryan also moved for reconsideration of the denial of Stokley’s post-conviction petition, and sought to amend the petition. Her proposed amended petition included a list of thirty-one new possible claims for relief. Ryan included a claim regarding the ineffectiveness of Levitt. She argued that “the substance of the Petition is deficient” and noted misstatements of law prejudicial to Stokley. Ryan specifically noted that she had not had an opportunity to do a full investigation, and that “other issues may need to be raised.”
After one month, Ryan was removed and Levitt was reinstated. Once reinstated, Levitt actively moved to defend herself and undermine Stokley’s case. Levitt systematically argued against the claims raised by Ryan. She noted that some were “already raised,” others “relate[d] to strategic decisions by the respective attorneys,” others were “contrary to well-established easelaw,” and still others were “not supported by the facts of the case.” Unexplainably, one of the claims Levitt derided as completely meritless was resurrected as the first of two additional claims in the supplemental Rule 32 petition. Thus, Levitt’s petition for review and later supplemental filing suggest an overriding concern with defending herself from the “attack on the effectiveness of undersigned counsel, all of which is meritless” rather than any loyal advocacy.
After Levitt was reinstated, Stokley wrote a letter to the Arizona Supreme Court asking for the reappointment of Ryan. This request was denied. Stokley then attempted to prepare his own claims and asked Levitt for a copy of the record. Levitt refused to give it to him. By failing to do so, she interfered with Stokley’s attempts to fairly present his claims.
The record shows that (1) both Stokley and his counsel agreed that their relationship had completely broken down; (2) Stokley took numerous steps to try to terminate the relationship and to obtain new counsel; (3) Levitt was reinstated as counsel over Stokley’s and her own objections; (4) Levitt was the subject of a Bar complaint; and (5) after she was reinstated as Stokley’s attorney, Levitt’s primary concern was to defend herself against misconduct charges. She disavowed and undermined the work Ryan had done on Stokley’s behalf, and refused Stokley access to his case file which limited his ability to marshal evidence and raise his own claims. Levitt ultimately came to the point where she was actively working against Stokley.
Stokley did everything in his power to sever his relationship with Levitt. The state vigorously advocated to make sure that Levitt was reinstated as his counsel. After the state prevailed, Levitt in effect worked in the state’s interest rather than in her client’s. As Stokley has argued before the district court and in the moving papers here, Levitt “took up the mantle of the prosecutor.” It is hard to imagine a clearer case for constructive abandonment.
The touchstone for understanding the Court’s decision in Maples is Justice Ali-to’s concurrence in Holland, . which the Court relies upon in explaining the meaning of “abandonment.” Holland, 130 S.Ct. at 2568. Justice Alito was not describing what happened in Stokley’s case. But he might as well have been.
II. Stokley’s colorable Eddings claim is sufficient prejudice to obtain remand.
Addressing prejudice at this stage is inconsistent with our prior precedent. *409Nevertheless, I feel compelled to respond to the majority’s argument.
The majority first states that, while Stokley’s causal nexus claim is colorable, the Arizona Supreme Court committed no actual error. This is incorrect. The majority goes on to assume that, even if the Arizona Supreme Court committed causal nexus error, the error was harmless. I address the second issue first, where the majority conflates structural and harmless error in a manner that confuses our prior case law and, without analysis, potentially closes an open and important question in the habeas law of our circuit.2 Whatever the ultimate outcome in Stokley’s case might have been had we remanded, by conflating structural and harmless error the majority creates tension with our prior case law and in my view sets a bad precedent.
Our prior cases have treated Eddings error as structural. We have consistently reversed and remanded Eddings cases to the Arizona courts for resentencing, without inquiring as to the likelihood of a different sentencing result. See, e.g., Williams v. Ryan, 623 F.3d 1258 (9th Cir.2010); Styers v. Schriro, 547 F.3d 1026 (9th Cir.2008). If an Eddings error is structural, as our cases suggest, prejudice is per se.
Citing Hitchcock v. Dugger, 481 U.S. 393, 399, 107 S.Ct. 1821, 95 L.Ed.2d 347 (1987), the panel concludes that Eddings errors are subject to harmless error review under Brecht v. Abrahamson, 507 U.S. 619, 113 S.Ct. 1710, 123 L.Ed.2d 353 (1993). Even assuming Eddings error is nonstructural, the panel appears to have erred in applying Brecht here because the state did not argue harmlessness in this court (until its response to the petition for rehearing), an issue on which the state bears the burden. See Hitchcock, 481 U.S. at 399, 107 S.Ct. 1821 (“Respondent has made no attempt to argue that this error was harmless, or that it had no effect on the jury or the sentencing judge. In the absence of such a showing our cases hold that the exclusion of mitigating evidence of the sort at issue here renders the death sentence invalid.”). As best I can tell, after finding Eddings error on habeas review, we have never engaged in harmless error review of the sort engaged in here.
Turning back to the majority’s finding that no Eddings violation occurred, I am unpersuaded by the panel’s analysis. Here, the Arizona Supreme Court did precisely what the Eighth Amendment prohibits—-it treated mitigating evidence of Stokley’s abusive childhood as nonmitigat-ing as a matter of law merely because it lacked a causal connection to the crime. The state court said:
According to a clinical psychologist, defendant had a chaotic and abusive childhood, never knowing his father and having been raised by various family members. A difficult family background alone is not a mitigating circumstance. State v. Wallace, 160 Ariz. *410424, 773 P.2d 983, 986 (1989), cert. denied, 494 U.S. 1047, 110 S.Ct. 1513, 108 L.Ed.2d 649 (1990). This can be a mitigating circumstance only “if a defendant can show that something in that background had an effect or impact on his behavior that was beyond the defendant’s control.” Id. ... Although [Stokley] may have had a difficult childhood and family life, he failed to show how this influenced his behavior on the night of the crimes.
State v. Stokley, 182 Ariz. 505, 898 P.2d 454, 473 (1995) (emphasis added).
This is a clear-cut Eddings violation, and the panel majority’s failure to recognize it cannot be squared with circuit precedent. We cannot avoid finding an Eddings violation, as the panel majority suggests, merely because the Arizona Supreme Court said it considered all mitigating evidence. See Styers, 547 F.3d at 1035. When a state court “considers” mitigating evidence, but deems it irrelevant or nonmiti-gating as a matter of law because of the absence of a causal connection to the crime, the court has not considered the evidence in any meaningful sense. See Penry v. Lynaugh, 492 U.S. 302, 319, 109 S.Ct. 2934, 106 L.Ed.2d 256 (1989), abrogated on other grounds by Atkins v. Virginia, 536 U.S. 304, 122 S.Ct. 2242, 153 L.Ed.2d 335 (2002).
Unlike the majority I would not reach the issues of either prejudice with respect to procedural default or the merits of the constitutional claim at this stage. When first presented with this claim that the Arizona Supreme Court erred in its review of the death sentence under Eddings and Skipper, the district court declined to reach the merits because the claim was technically exhausted and procedurally barred. Case 4:98-cv-00332-FRZ, Dkt. 70, Order and Opinion on Procedural Status of Claims at 15-16. No court has considered the issue of prejudice—either as to procedural default or to the merits of the constitutional claim—because, prior to Maples, there was no cause for the procedural default. Coleman v. Thompson, 501 U.S. 722, 750, 111 S.Ct. 2546, 115 L.Ed.2d 640 (1991). All that is required for prejudice at this stage is that the claim has some merit. Martinez v. Ryan, — U.S. -, 132 S.Ct. 1309, 1318, 182 L.Ed.2d 272 (2012).
Without the benefit of any briefing or lower court consideration on the issue of prejudice arising from the defaulted Ed-dings and Skipper claims, we are not in a position to do what the majority does here. Rather than foreclosing these claims at this stage, I would stay the mandate and remand this case to the district court for the limited purpose of allowing it to determine in the first instance whether cause and prejudice exist, and to consider the merits of the claim if warranted. We would then be in a far better position to review the issue.
For all of the above reasons, I respectfully dissent.
. I agree with the majority’s assumption that Maples may be sufficient to establish the "exceptional circumstance" necessary to justify the exercise of this court’s power to stay the mandate following a denial of certiorari.
. As I understand it, the Supreme Court has not addressed whether Eddings error is structural nor has this court squarely examined the issue. Compare Landrigan v. Stewart, 272 F.3d 1221, 1230 & n. 9 (9th Cir.2001) (applying harmless error review to the state court’s failure to consider the defendant's alleged intoxication and past history of drug use as a nonstatutory mitigating factor), adopted by Landrigan v. Schriro, 501 F.3d 1147, 1147 (9th Cir.2007) (en banc) (order), with Williams V. Ryan, 623 F.3d 1258, 1270-71 (9th Cir.2010) (granting habeas relief for an Eddings violation without conducting a harmless error analysis), and Styers v. Schriro, 547 F.3d 1026, 1035-36 (9th Cir.2008) (same). Other circuits are split on the issue. Compare Bryson v. Ward, 187 F.3d 1193, 1205 (10th Cir.1999) (collecting cases applying harmless error review), with Nelson v. Quarterman, 472 F.3d 287, 314 (5th Cir.2006) (en banc) (declining to apply harmless error review).